# 项目论文真机机器人型号调研：WAM 与 VLA

本调研以项目内论文原文为范围，整理实际使用的机器人、具体型号、双臂／移动／灵巧手配置，并比较 WAM 与 VLA 的使用方式。**两类方法大量共用硬件，主要差别是如何学习、预测和执行动作，而不是机器人品牌。**

范围快照：2026-10-09。时间不另设起止条件；项目里所有已发现论文均进入全量审计表。arXiv ID 的月份范围为 **2025-08 至 2026-10**，不等同下载版本的发表日期或会议年份；旧 ID 的新版本同样纳入。这里只统计本地语料，不宣称覆盖该时期整个研究领域，也未独立验证论文实验真实性。

## 如何查阅

- 本页：统计口径、硬件型号总览、WAM/VLA 对比、逐篇真机清单。
- [全部 855 篇审计表](research/robot_hardware/all_papers.md)：包含未纳入真机统计的论文。
- [CSV 数据表](research/robot_hardware/all_papers.csv)：可按目录、状态、型号筛选；含本地 PDF／文本／阅读报告路径。
- [原文定位摘录](research/robot_hardware/evidence.md)：正文和附录的检索片段、对应文本行号及原文链接。
- [结构化审计记录](research/robot_hardware/papers.json)：逐篇结论；[生成脚本](research/robot_hardware/build_readme.py)可重建表格与统计。

## 启动交互式可视化

在项目根目录运行（仅需 Python 3，无需前端依赖）：

```bash
python3 robot-dashboard/serve.py --port 8000
```

打开 http://localhost:8000 。若项目运行在远程服务器，可通过 SSH 转发端口：`ssh -N -L 8000:127.0.0.1:8000 用户名@服务器地址`，再在自己的浏览器打开该地址。服务需保持运行；连接中断后可重新执行启动命令。

页面提供机器人家族分布、原始热度概览、型号与构型加权排名、论文搜索和证据详情。点击家族条形或型号名称可联动筛选论文；型号 CSV 导出当前筛选结果。热度为已保存快照，缺失项单独标明。不要直接双击 HTML，数据与本地原文需要通过上述服务加载。

## 域名部署

域名：`robot.bizziam.com`。DNS 添加 A 记录 `robot → 50.114.5.104`（默认 TTL；首次签发证书建议使用仅 DNS 模式）。HTTPS 已于 2026-10-09 启用，HTTP 自动跳转至 HTTPS；证书自动续期定时器已启用。以下命令用于重新申请／安装该域名证书：

```bash
bash robot-dashboard/deploy/enable-https.sh
```

线上目录为 `/var/www/robot.bizziam.com/current`，仅发布图谱及白名单原文文件，不依赖本地 Python 服务。更新页面或重建数据后，运行 `python3 robot-dashboard/deploy/publish.py` 原子切换发布快照。PDF／文本优先使用硬链接节省空间；旧快照保留用于回退。Nginx 配置位于 `robot-dashboard/deploy/`，证书续期沿用服务器现有 Certbot 设置。

## 范围、去重和判断标准

扫描 `wam论文`、`vla论文` 下的 PDF、提取文本和阅读报告，再按 arXiv ID 去重。正文／附录的实验设置优先于阅读报告；报告只辅助定位。检索覆盖完整提取文本，不只搜索标题、摘要或已知机器人品牌；对实体实验候选核对实验描述、结果、图注及附录。型号未披露或本轮不能可靠定位时保留“未明”，不根据图片外观、训练数据来源或相似实验替作者补型号。

| 范围 | PDF 文件数（去重前） | 去重论文数 | 说明 |
|---|---:|---:|---|
| WAM 目录 | 236 | 237 | 有仅文本／报告记录 |
| VLA 目录 | 665 | 654 | 同 ID 可有重复缓存／报告 |
| 两目录并集 | 901 | 855 | 36 篇同时收录；853 篇有 PDF，2 篇仅文本／报告 |

另有 **73 篇只有缓存原文而没有对应标题／阅读报告条目**，本轮从原文补齐标题并纳入，不以阅读报告数量作为分母。

这里的 WAM 指 **World Action Model**，不是 Barrett WAM 机械臂。`WAM`、`VLA` 标签首先表示项目目录归属；双收录标为 `WAM/VLA`。目录不是互斥学术分类，含世界预测模块的 VLA、使用 VLA 作为动作专家的 WAM、两类方法共同使用的基准都可能跨界。

| 状态 | 本调研的含义 | 是否计入主要真机统计 |
|---|---|---|
| R | 本文报告实体机器人实际运行，包括闭环策略、在线学习、定性执行、开环执行或实际系统实验；不是一律等价的实验强度 | 是 |
| D | 单列识别出的真实数据分析、观测回放、采集或硬件时延测量，未确认本文策略实体部署 | 否，另列 |
| U | 真机证据不足、表述有歧义或尚待报告结果 | 否，另列 |
| N | 本轮没有确认新增实体运行；包括仿真、综述、既有数据评估等 | 否，见全量表 |

**R 是本轮识别结果，不是“所有论文均逐页人工验证”的承诺。** 全文检索和候选人工复核仍可能漏掉只在图片中披露的型号。D 只单列本轮辨认出的相关边界项目，并不穷举所有 OXE、DROID、Bridge、驾驶日志等数据使用；N 不能解读成“没有真实数据”。统计反映论文报告的实验，不将运行次数、机器人台数、型号配置数混为论文数。

## 本地语料统计

| 归属 | 全部论文 | R：实体运行 | D | U | N |
|---|---:|---:|---:|---:|---:|
| WAM | 237 | 167 | 3 | 1 | 66 |
| VLA | 654 | 436 | 6 | 3 | 209 |
| 去重并集 | 855 | 575 | 9 | 4 | 267 |

两目录计数重叠，不能相加当作全局篇数。一篇使用多个品牌，会在多个硬件家族中出现；同篇使用某家族多台机器人，家族论文数只计一次。以下家族表**用于检索，不是精确型号排行榜**；例如 DROID 型号未核者可进入 Franka/DROID 家族，不能因此计为 Panda。

## 机器人家族与型号总览

新增：[型号与构型热度加权排名](research/robot_hardware/model_counts.md) · [型号统计 CSV](research/robot_hardware/model_counts.csv) · [逐篇热度权重 CSV](research/robot_hardware/paper_weights.csv)。按实体运行论文去重计数，保留普通篇数，并增加单臂、双臂、单臂＋轮式底盘、双臂＋轮式底盘及灵巧手标签。标签可重叠；无法确定的标为待确认。

热度沿用原门户：`max(GitHub Stars, Hugging Face 热度)`，直接作为论文权重，不分档、不归一化。有热度的 **221 篇**使用原始值；**634 篇**缺失项权重为空，不参与热度求和，仍计入论文篇数。已知零热度与缺失值分开。同热度权重相同，更新时间和 arXiv ID 仅用于展示排序。型号得分为相关 R 类去重论文的已知原始热度之和。快照日期为 2026-09-16／17，非实时热度。

构型按“论文 × 型号”标注，查看页面详情可追溯分类依据；整机默认双臂构型不表示策略同时控制所有关节。运行 `python3 research/robot_hardware/count_models.py` 可从已保存热度快照重建分类、权重和型号表。

| 家族／平台 | R 篇数（去重） | WAM目录 | VLA目录 | 部分论文入口 |
|---|---:|---:|---:|---|
| Franka | 106 | 27 | 83 | [2607.02195](#paper-2607.02195)、[2607.02840](#paper-2607.02840)、[2607.02865](#paper-2607.02865)、[2607.03387](#paper-2607.03387) |
| AgileX PiPER / PiPER-X / PiPER-H | 93 | 26 | 72 | [2601.17885](#paper-2601.17885)、[2607.00678](#paper-2607.00678)、[2607.01804](#paper-2607.01804)、[2607.12356](#paper-2607.12356) |
| AgileX Cobot / Cobot Magic | 24 | 8 | 17 | [2607.06403](#paper-2607.06403)、[2607.12931](#paper-2607.12931)、[2607.13597](#paper-2607.13597)、[2607.24008](#paper-2607.24008) |
| ALOHA / Mobile ALOHA | 23 | 8 | 17 | [2607.03693](#paper-2607.03693)、[2607.04171](#paper-2607.04171)、[2607.04265](#paper-2607.04265)、[2607.04927](#paper-2607.04927) |
| Universal Robots UR | 49 | 11 | 40 | [2607.00666](#paper-2607.00666)、[2607.04816](#paper-2607.04816)、[2607.08182](#paper-2607.08182)、[2607.08974](#paper-2607.08974) |
| UFACTORY xArm / UF850 | 29 | 6 | 26 | [2607.02503](#paper-2607.02503)、[2607.13429](#paper-2607.13429)、[2607.14852](#paper-2607.14852)、[2608.04633](#paper-2608.04633) |
| ARX | 29 | 12 | 18 | [2607.05377](#paper-2607.05377)、[2607.05468](#paper-2607.05468)、[2607.06403](#paper-2607.06403)、[2607.08436](#paper-2607.08436) |
| SO-100 / SO-101 / SO-ARM101 | 29 | 2 | 27 | [2607.02322](#paper-2607.02322)、[2607.04146](#paper-2607.04146)、[2607.04591](#paper-2607.04591)、[2607.06370](#paper-2607.06370) |
| Unitree 人形 G1 | 21 | 10 | 11 | [2607.06988](#paper-2607.06988)、[2607.16636](#paper-2607.16636)、[2607.18016](#paper-2607.18016)、[2607.20345](#paper-2607.20345) |
| Unitree 四足 / Z1 / D1 | 8 | 0 | 8 | [2607.08575](#paper-2607.08575)、[2607.16636](#paper-2607.16636)、[2608.10756](#paper-2608.10756)、[2608.17717](#paper-2608.17717) |
| AgiBot | 17 | 5 | 13 | [2607.18231](#paper-2607.18231)、[2607.28405](#paper-2607.28405)、[2608.03483](#paper-2608.03483)、[2608.05738](#paper-2608.05738) |
| Astribot | 10 | 5 | 6 | [2607.06403](#paper-2607.06403)、[2607.11270](#paper-2607.11270)、[2608.14047](#paper-2608.14047)、[2608.22067](#paper-2608.22067) |
| Galaxea | 10 | 4 | 8 | [2603.16666](#paper-2603.16666)、[2607.08283](#paper-2607.08283)、[2608.09730](#paper-2608.09730)、[2609.03681](#paper-2609.03681) |
| RealMan | 9 | 1 | 8 | [2608.06434](#paper-2608.06434)、[2608.06965](#paper-2608.06965)、[2608.09125](#paper-2608.09125)、[2608.15816](#paper-2608.15816) |
| I2RT YAM | 12 | 8 | 7 | [2607.08127](#paper-2607.08127)、[2608.10232](#paper-2608.10232)、[2608.10860](#paper-2608.10860)、[2609.20648](#paper-2609.20648) |
| DOBOT | 9 | 2 | 7 | [2607.02195](#paper-2607.02195)、[2607.16636](#paper-2607.16636)、[2608.05042](#paper-2608.05042)、[2608.25395](#paper-2608.25395) |
| Flexiv | 6 | 4 | 3 | [2607.14236](#paper-2607.14236)、[2607.23783](#paper-2607.23783)、[2609.05266](#paper-2609.05266)、[2609.20761](#paper-2609.20761) |
| Tianji / Marvin | 11 | 8 | 3 | [2607.12287](#paper-2607.12287)、[2608.22403](#paper-2608.22403)、[2608.25757](#paper-2608.25757)、[2608.27033](#paper-2608.27033) |
| TienKung / TianGong | 3 | 1 | 2 | [2607.06655](#paper-2607.06655)、[2608.16837](#paper-2608.16837)、[2609.18197](#paper-2609.18197) |
| ROKAE | 3 | 1 | 2 | [2608.26578](#paper-2608.26578)、[2609.37530](#paper-2609.37530)、[2609.40219](#paper-2609.40219) |
| AIRBOT | 4 | 1 | 3 | [2607.02322](#paper-2607.02322)、[2608.16172](#paper-2608.16172)、[2609.23565](#paper-2609.23565)、[2610.01698](#paper-2610.01698) |
| WidowX / Interbotix | 6 | 0 | 6 | [2508.13446](#paper-2508.13446)、[2607.02466](#paper-2607.02466)、[2607.03449](#paper-2607.03449)、[2608.02958](#paper-2608.02958) |
| Rainbow Robotics RB-Y1 | 3 | 0 | 3 | [2609.26672](#paper-2609.26672)、[2610.00913](#paper-2610.00913)、[2610.10437](#paper-2610.10437) |
| OpenArm | 3 | 2 | 1 | [2610.02840](#paper-2610.02840)、[2610.05483](#paper-2610.05483)、[2610.09696](#paper-2610.09696) |
| Synria Alicia | 3 | 0 | 3 | [2608.07314](#paper-2608.07314)、[2608.10756](#paper-2608.10756)、[2609.07006](#paper-2609.07006) |
| Galbot | 3 | 3 | 0 | [2607.06988](#paper-2607.06988)、[2609.15570](#paper-2609.15570)、[2609.24868](#paper-2609.24868) |
| LimX | 2 | 1 | 1 | [2609.18108](#paper-2609.18108)、[2610.06965](#paper-2610.06965) |
| Kinova | 2 | 0 | 2 | [2607.01212](#paper-2607.01212)、[2609.03142](#paper-2609.03142) |
| ROBOTIS OMY | 2 | 0 | 2 | [2610.05026](#paper-2610.05026)、[2610.08425](#paper-2610.08425) |
| 其他工业／协作平台 | 6 | 1 | 5 | [2607.06262](#paper-2607.06262)、[2608.27079](#paper-2608.27079)、[2609.03142](#paper-2609.03142)、[2609.17115](#paper-2609.17115) |
| 移动导航／专用平台 | 34 | 10 | 24 | [2508.13446](#paper-2508.13446)、[2607.03146](#paper-2607.03146)、[2607.10180](#paper-2607.10180)、[2607.12659](#paper-2607.12659) |

### 具体型号与配置要点

| 类别 | 本地论文中记录的型号／配置 | 阅读时要保留的区别 |
|---|---|---|
| Franka | Panda／Emika Panda、Research 3／FR3、FR3 Duo、单臂与双臂 | 只写 Franka 的论文不细分 Panda/FR3；FR3 Duo 视为系统配置 |
| AgileX | PiPER、PiPER-X、PiPER-H、NERO；Cobot／Cobot Magic／V2；PiPER＋RANGER MINI 3／TRACER | Cobot Magic 是平台，执行臂和底盘按原文分别记录；PiPER 12 保留原文命名，不推定独立商品型号 |
| ALOHA 系列 | ALOHA、ALOHA 2、Mobile ALOHA、Trossen Stationary AI Kit、AgileX ALOHA、AIRBOT Mobile ALOHA | 名称相似不保证使用相同机械臂；ALOHA-style 仅表示构型相似 |
| UR | UR3、UR3e、UR5、UR5e、UR7e、UR10、UR10e；UR AI Trainer | e 系列不可与非 e 系列合并成精确型号；双臂另注明 |
| UFACTORY | xArm6、xArm7、UF850／论文写法 xArm 850；未指明代际的 xArm | 保留原文不同命名，不擅自把所有 xArm 统一成 7 轴 |
| ARX | ARX 5／ARX-5、R5、X5、ACONE／AC-One／AC1、LIFT／LIFT2／Lift-2S | 未查清官方对应关系的近似名字不强行归并；主臂与从臂角色不同 |
| 低成本及开放平台 | SO-100、SO-101、SO-ARM101、SO-101 Arm Pro、YAM／YAM Box、OpenArm V1、OMY／OMY F3M | 夹爪算动作维度不等于机械臂关节数；左右臂及执行端单列 |
| 其他协作／工业臂 | WidowX-250s、AIRBOT Play、Kinova Gen3、Flexiv Rizon 4／4R／4S、Dobot Nova2／Nova／CR3／CR5A、RealMan RM65-B／RM-65B／RM75／RM75-6F／Gen72、ROKAE AR5／AR5-5 0.7、KUKA iiwa 14、JAKA mini2、COMAU Racer 3、FANUC LR Mate 200iD、MELFA ASSISTA、Synria Alicia-D | 家族名不是型号；后缀不明就保留，不依据自由度推断 |
| 人形与轮式双臂 | Unitree G1／G1-Edu／G1D／G1-D／G1-comp；AgiBot G1／G2／A2／A3／A3 Ultra；Astribot S1；Galaxea A1／R1 Lite／R1 Pro；Galbot G1；RB-Y1；TienKung 2.0／3.0、TianGong 3.0；Tianji Marvin／Marvin Pro；LimX TRON2 | Unitree G1、AgiBot G1、Galbot G1 是不同平台；裸写 G1 不代填厂商；人形上身操作不等于全身行走控制 |
| 其他整机／专用平台 | Spirit AI MOZ1、IMETA-Y1、PrimeBot X2-W、Leju Kuavo、Loong S1、KEENON XMAN-R1、XLeRobot、LingYu TeleAvatar、Toyota HSR、XSquare Robot、NICOL、ALLEX、NCS 系列、Omega 等 | 详见逐篇清单；保留作者系统名，不把系统名强行变成商业型号 |
| 导航与移动操作 | Kobuki、Unitree Go2／Go2 EDU、B2＋Z1、D1-T、Boston Dynamics Spot、Clearpath Jackal、DeepRobotics M20S＋CM1 | 单纯导航和带臂移动操作分开看；底盘与上装应分别记录 |
| 非常规实体 | XCMG XC958EV 装载机、自制四旋翼（含 280 mm、OddityRC 35Pro 机架）、水下机器人、内镜／医疗操作平台、软体机器人 | 自制机架名不是整机型号；汽车离线日志不是实车部署 |
| 末端与触觉 | Robotiq 2F-140、PGI-140-80、LinkerHand L20、Inspire RH56F1、Wuji／Wuji Hand 2、SharpaWave、XHand／XHand1、BrainCo Revo2、Dex1-1、CRAFT、YUBI、Pika、Shadow Dexterous Hand Lite；GelSight／Xense／NeoteAI／DM-TacW2 | 手、夹爪、触觉传感器与机械臂属于不同层级；夹爪品牌不能证明机械臂品牌 |

型号表汇总的是下方逐篇记录；若同一写法来自数据／未决条目，则以该条目的状态为准，不自动计入 R。

官方资料仅辅助解释命名，不替代论文的使用证据：[Trossen ALOHA 规格](https://docs.trossenrobotics.com/aloha_docs/1.0/specifications.html)列出原版从臂 ViperX 300S、主臂 WidowX 250S；这不能直接套用到每篇自称 ALOHA 的平台。[Interbotix 机械臂文档](https://docs.trossenrobotics.com/interbotix_xsarms_docs/)用于核对产品命名；[Franka Research 3 官方页面](https://franka.de/franka-research-3-arm)用于区分 FR3。未针对每篇论文联网追踪最新版本，结论以下载版本为准。

## WAM 与 VLA 使用真机的异同

| 对比项 | WAM 在本地论文中的典型用途 | VLA 在本地论文中的典型用途 | 对硬件选择／实验解读的影响 |
|---|---|---|---|
| 输入和输出 | 未来图像、状态、光流与动作联合建模；通过动作专家、逆动力学或规划输出控制 | 图像、语言、本体状态到动作；连续动作块、离散 token、flow／diffusion 等 | 两者都能控制同一 Franka、PiPER、ARX 或 YAM，不存在硬件上的严格分界 |
| 真机验证重点 | 想象与实际执行的一致性、动态目标、反事实规划、长程预测、模型与现实偏差 | 指令跟随、泛化、记忆、执行精度、纠错、在线 RL、部署加速 | 用同一机器人也可能验证完全不同的能力，成功率不可跨论文直接排名 |
| 时延与控制 | 视觉未来生成常增加推理负担，研究异步预测、少步生成和执行期间重规划 | 视觉 token、动作解码、缓存、量化、异步动作块优化同样重要 | 要区分主机推理吞吐、策略更新频率、动作执行频率和底层伺服频率 |
| 传感器 | 常见主视角＋腕视角；预测多视角、深度、光流或潜在状态 | 同样普遍使用主／腕相机；部分引入力、触觉、雷达、眼动等 | 多相机不是 WAM 独有；RGB-D 相机存在不代表模型真的消费深度 |
| 双臂与灵巧手 | 用于联合时空预测、长程协作和高维动作；Tianji＋Wuji／Sharpa 等 | 用于动作表示、协调、示教迁移、触觉适应和在线学习 | 应同时记录机械臂、末端、底盘；不能把所有双臂叫作 ALOHA |
| 导航和移动 | 世界预测、路线推演、移动操作与操作前停靠 | 指令导航、跨本体迁移、导航与抓取衔接、全身控制 | 需要区分地面移动、人形行走、移动操作与纯桌面操作 |
| 真机证据强度 | 真实视频预测可能完全离线；世界模型生成的 rollout 不一定是实体执行 | “real-to-sim”、真实数据训练或仿真中的 Panda 也不代表真机部署 | 必须找本文物理平台、实际运行结果和部署描述，不能只看 real-world 关键词 |

### 可以直接对照的论文

- **FlowWAM**：Franka FR3 与 ARX 5 双臂；附录明确平台、相机和控制栈。光流／未来预测最终仍需转换成实体动作。
- **Flex-π、FACT、Dynamic Manipulation via Counterfactual Planning**：使用 YAM 双臂。**Action Upcycling、GT-VLA、FastOPD** 同样使用 YAM，体现共享硬件而研究问题不同。
- **VLA-Precision**：UR5e 的平行夹爪、LinkerHand L20、双 UR5e，以及 Franka FR3；适合看同一方法如何跨末端和单双臂验证。
- **UniWAM**：Cobot Magic、AgiBot G2、双 FR3、AgileX 单臂；导航、移动操作、固定操作分别使用哪些平台要看任务映射。
- **FLEX-WAM**：OpenArm V1 双臂有真实闭环部署；Unitree G1 的长程生成属于真实数据上的预测，不能把它一起算作 G1 实体部署。
- **OpenViTac**：WAM、VLA、VTLA 在成对 sim/real 任务上比较，提供更可比的任务与感知条件；原文可确认的夹爪、触觉配置不等于已经确认机械臂型号。

这些例子的完整标题、ID 和本地来源均在下方逐篇表中。比较结论是对本地语料的归纳，不表示全部 WAM 或全部 VLA 必然如此。

## 容易误报、漏报的情况

1. **LIBERO／RLBench／RoboTwin／SimplerEnv 里的 Panda、UR、WidowX**：先确认是否为仿真本体；真实数据训练后仅跑仿真不计 R。
2. **DROID 不等于本篇用了 Panda**：ProWAM 明确使用 FR3 并沿用 DROID 配置；只提 DROID 检查点的论文不由此推断硬件。
3. **ALOHA 是构型／系统名**：原版、ALOHA 2、Mobile ALOHA、AgileX 和 AIRBOT 变体分开；遥操作主臂不当作执行从臂。
4. **仅报告未来计划**：SeeReach-VLA 的下载版本仍标记结果待报告，保留 U；LaMem-VLA 明确将真机留到下一版本，不计 R。
5. **阅读报告可能落后于原文**：PonderPounce 的附录 E 包含 Trossen ALOHA 真机试验；本轮以原文补入。
6. **只写 Franka、G1、ARX 或双臂机器人**：保留能够证明的粒度，未核实的型号不自动细化。未决项和型号未明项都保留，避免以“无法确定型号”为由漏掉真实实验。

## 逐篇真机清单（R）

每行是一篇去重论文。点击标题打开本地 PDF（缺 PDF 时为文本），点击“证据”查看原文定位；WAM/VLA 双收录只列一次。机器人型号字段保留不确定性；末端或主从角色在备注中说明。

| ID／目录 | 论文（本地原文） | 真机型号／配置 | 用途与辨析 |
|---|---|---|---|
| <a id="paper-2508.13446"></a>2508.13446<br>VLA | [CAST: Counterfactual Labels Improve Instruction Following in Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2508.13446.pdf) | Kobuki 移动底盘；WidowX（版本未明） | 导航与操作使用不同实体；[证据](research/robot_hardware/evidence.md#paper-2508.13446) |
| <a id="paper-2601.17885"></a>2601.17885<br>VLA | [PEAfowl: Perception-Enhanced Multi-View Vision-Language-Action for Bimanual Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2601.17885.pdf) | AgileX PiPER | 真机；ALOHA-AgileX 为仿真；[证据](research/robot_hardware/evidence.md#paper-2601.17885) |
| <a id="paper-2603.16666"></a>2603.16666<br>WAM | [Fast-WAM: Do World Action Models Need Test-time Future Imagination?](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2603.16666.pdf) | Galaxea R1 Lite | 叠毛巾；[证据](research/robot_hardware/evidence.md#paper-2603.16666) |
| <a id="paper-2607.00666"></a>2607.00666<br>VLA | [Domain Arithmetic: One-Shot VLA Adaptation under Environmental Shifts](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.00666.pdf) | UR10e | Panda→UR5e 是仿真跨本体；[证据](research/robot_hardware/evidence.md#paper-2607.00666) |
| <a id="paper-2607.00678"></a>2607.00678<br>WAM | [ABot-M0.5: Unified Mobility-and-Manipulation World Action Model](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.00678.pdf) | AgileX PiPER | 单臂定量；移动平台型号未明；[证据](research/robot_hardware/evidence.md#paper-2607.00678) |
| <a id="paper-2607.01212"></a>2607.01212<br>VLA | [FurnitureVLA: Learning Long-Horizon Bimanual Furniture Assembly with Vision-Language-Action Model](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.01212.pdf) | Kinova Gen3 ×2 | 左 Robotiq Hand-E；右 2F-85；[证据](research/robot_hardware/evidence.md#paper-2607.01212) |
| <a id="paper-2607.01804"></a>2607.01804<br>VLA | [VLA-Corrector: Lightweight Detect-and-Correct Inference for Adaptive Action Horizon](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.01804.pdf) | AgileX PiPER | 6-DoF；[证据](research/robot_hardware/evidence.md#paper-2607.01804) |
| <a id="paper-2607.02092"></a>2607.02092<br>VLA | [Guided Action Flow: Q-Guided Inference for Flow-Matching Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.02092.pdf) | 未公布型号 | 内部平台；[证据](research/robot_hardware/evidence.md#paper-2607.02092) |
| <a id="paper-2607.02195"></a>2607.02195<br>WAM | [Bridge-WA: Predicting Where and How the World Changes for Robotic Action](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.02195.pdf) | Dobot Nova2；Franka FR3 | 原文附录补充了 FR3；阅读报告只提 Dobot；[证据](research/robot_hardware/evidence.md#paper-2607.02195) |
| <a id="paper-2607.02322"></a>2607.02322<br>VLA | [The Moving Eye: Enhancing VLA Spatial Generalization via Hybrid Dynamic Data Collection](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.02322.pdf) | SO-101；AIRBOT（版本未明） | SO-101 操作；AIRBOT 移动相机；[证据](research/robot_hardware/evidence.md#paper-2607.02322) |
| <a id="paper-2607.02466"></a>2607.02466<br>VLA | [Learning to Move Before Learning to Do: Task-Agnostic pretraining for VLAs](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.02466.pdf) | WidowX-250s | 自主 play 与两项闭环任务；[证据](research/robot_hardware/evidence.md#paper-2607.02466) |
| <a id="paper-2607.02503"></a>2607.02503<br>WAM | [VT-WAM: Visual-Tactile World Action Model for Contact-Rich Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.02503.pdf) | xArm7 | Robotiq 2F-85；双 Xense；[证据](research/robot_hardware/evidence.md#paper-2607.02503) |
| <a id="paper-2607.02840"></a>2607.02840<br>VLA | [TACO: TActile World Model as a Self-COrrector forScalable VLA Post-Training](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.02840.pdf) | Franka FR3 | Xense 触觉；[证据](research/robot_hardware/evidence.md#paper-2607.02840) |
| <a id="paper-2607.02865"></a>2607.02865<br>VLA | [DREAMSTEER: Latent World Models Can Steer VLA Policies During Deployment Without Any Finetuning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.02865.pdf) | Franka Panda | 真实控制；[证据](research/robot_hardware/evidence.md#paper-2607.02865) |
| <a id="paper-2607.03146"></a>2607.03146<br>VLA | [Exp2VLA: Enabling Vision-Language-Action for Drone Navigation from Expert Demonstrations](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.03146.pdf) | 自制四旋翼（型号未明） | 实飞演示；SITL 另列；[证据](research/robot_hardware/evidence.md#paper-2607.03146) |
| <a id="paper-2607.03387"></a>2607.03387<br>VLA | [Feeling the Unexpected: ResTacVLA for Contact-Rich Manipulation via Residual Tactile Representation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.03387.pdf) | Franka FR3 | Robotiq 2F-85；GelSight Mini；[证据](research/robot_hardware/evidence.md#paper-2607.03387) |
| <a id="paper-2607.03449"></a>2607.03449<br>VLA | [HiMe: Hierarchical Embodied Memory for Long-Horizon Vision-Language-Action Control](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.03449.pdf) | WidowX-250s | 正文亦写 WidowX-250；[证据](research/robot_hardware/evidence.md#paper-2607.03449) |
| <a id="paper-2607.03693"></a>2607.03693<br>VLA | [CoRE-VLA: Towards Scalable and Robust Vision-Language-Action Modeling via Conditional Routing of Experts](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.03693.pdf) | AgileX ALOHA（臂型号未明） | 双臂；深度模态；[证据](research/robot_hardware/evidence.md#paper-2607.03693) |
| <a id="paper-2607.04146"></a>2607.04146<br>VLA | [!Imperio, smolVLA: The Implications of Data Poisoning on Open Source Robotics](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.04146.pdf) | SO-101 | 物理抓放实验；[证据](research/robot_hardware/evidence.md#paper-2607.04146) |
| <a id="paper-2607.04171"></a>2607.04171<br>VLA | [Teaching Tiny VLA Models Where to Look and How to Move](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.04171.pdf) | Mobile ALOHA（臂型号未明） | 不据系统名反推执行臂；[证据](research/robot_hardware/evidence.md#paper-2607.04171) |
| <a id="paper-2607.04265"></a>2607.04265<br>WAM | [HALO-WA: Hybrid-Attention Latent-Guided Online Reinforcement Learning for World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.04265.pdf) | AgileX ALOHA（臂型号未明） | Songling/松灵与 AgileX 品牌别名；[证据](research/robot_hardware/evidence.md#paper-2607.04265) |
| <a id="paper-2607.04591"></a>2607.04591<br>VLA | [Simple-to-Complex Structured Demonstrations for Vision-Language-Action Learning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.04591.pdf) | SO-101 ×2 | 双臂四相机；[证据](research/robot_hardware/evidence.md#paper-2607.04591) |
| <a id="paper-2607.04816"></a>2607.04816<br>VLA | [CAC-VLA: Context-Gated Action Conditioning for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.04816.pdf) | UR7e | 抓放及叠块；[证据](research/robot_hardware/evidence.md#paper-2607.04816) |
| <a id="paper-2607.04880"></a>2607.04880<br>VLA | [PRISM: Personalized Robotic Dataset Generation via Image-based Scene and Motion Synthesis](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.04880.pdf) | Franka（版本未明） | 合成演示迁移真机；[证据](research/robot_hardware/evidence.md#paper-2607.04880) |
| <a id="paper-2607.04927"></a>2607.04927<br>WAM | [DSWAM: A Dual-System World Action Foundation Model for Fine-Grained Robot Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.04927.pdf) | ALOHA-style（厂商及臂型号未明） | 衣物折叠；[证据](research/robot_hardware/evidence.md#paper-2607.04927) |
| <a id="paper-2607.05377"></a>2607.05377<br>VLA | [Cortex: A Bidirectionally Aligned Embodied Agent Framework for Long-horizon Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.05377.pdf) | ARX ACONE | 长程化学实验；[证据](research/robot_hardware/evidence.md#paper-2607.05377) |
| <a id="paper-2607.05396"></a>2607.05396<br>VLA | [From Fixed to Free Cameras: Calibration-Free View-Robust Vision-Language-Action Model](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.05396.pdf) | Franka Research 3（FR3） | 真实控制；[证据](research/robot_hardware/evidence.md#paper-2607.05396) |
| <a id="paper-2607.05468"></a>2607.05468<br>WAM | [Learning 4D Geometric Priors for Inference-Efficient World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.05468.pdf) | ARX R5 | 码块与排序；[证据](research/robot_hardware/evidence.md#paper-2607.05468) |
| <a id="paper-2607.06262"></a>2607.06262<br>VLA | [Optimal Transport Q-Learning for Flow Policy Steering and Acceleration](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.06262.pdf) | KUKA iiwa 14 | Robotiq 2F-85；原文附录 E；[证据](research/robot_hardware/evidence.md#paper-2607.06262) |
| <a id="paper-2607.06370"></a>2607.06370<br>VLA | [ActionCache: Training-Free Acceleration for Vision-Language-Action Models with Action Caching and Refinement](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.06370.pdf) | SO-101 | 缓存加速；[证据](research/robot_hardware/evidence.md#paper-2607.06370) |
| <a id="paper-2607.06403"></a>2607.06403<br>VLA | [From Foundation to Application: Improving VLA Models in Practice](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.06403.pdf) | Astribot S1；Cobot Magic-ARX X5；AgileX Cobot Magic | 另有 GM-100 多本体，见证据补充；[证据](research/robot_hardware/evidence.md#paper-2607.06403) |
| <a id="paper-2607.06564"></a>2607.06564<br>VLA | [Lift3D-VLA: Lifting VLA Models to 3D Geometry and Dynamics-Aware Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.06564.pdf) | Franka FR3（单臂及双臂） | 单臂六任务与双臂任务；[证据](research/robot_hardware/evidence.md#paper-2607.06564) |
| <a id="paper-2607.06655"></a>2607.06655<br>VLA | [Pelican-VLA 0.5: Attending Before Acting Benefits Generalization](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.06655.pdf) | TienKung（版本未明） | 人形机器人；[证据](research/robot_hardware/evidence.md#paper-2607.06655) |
| <a id="paper-2607.06678"></a>2607.06678<br>VLA | [NativeMEM: Native Memory Compression for Long-Horizon Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.06678.pdf) | 型号未明 | 长程真机任务；[证据](research/robot_hardware/evidence.md#paper-2607.06678) |
| <a id="paper-2607.06988"></a>2607.06988<br>WAM | [WAM-TTT: Steering World-Action Models by Watching Human Play at Test Time](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.06988.pdf) | Unitree G1；Galbot（夹爪版）；Galbot（Sharpa 手版） | Unitree Dex3；Galbot 具体代际未明；[证据](research/robot_hardware/evidence.md#paper-2607.06988) |
| <a id="paper-2607.08127"></a>2607.08127<br>WAM | [Understanding and Mitigating the Video-Action Generalization Gap via Temporal Ratio](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.08127.pdf) | I2RT YAM ×2 | 双臂各6DoF；三台D405；[证据](research/robot_hardware/evidence.md#paper-2607.08127) |
| <a id="paper-2607.08182"></a>2607.08182<br>VLA | [LEEVLA: Seeing What Matters in Latent Environment Evolution for Vision-Language-Action](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.08182.pdf) | UR5 | 不要自动补 e 后缀；[证据](research/robot_hardware/evidence.md#paper-2607.08182) |
| <a id="paper-2607.08283"></a>2607.08283<br>VLA | [TFP: Temporally Conditioned Memory-Fusion Policies for Visuomotor Learning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.08283.pdf) | Galaxea A1 | 双臂任务；[证据](research/robot_hardware/evidence.md#paper-2607.08283) |
| <a id="paper-2607.08436"></a>2607.08436<br>WAM | [EgoWAM: World Action Models Beyond Pixels with In-the-Wild Egocentric Human Data](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.08436.pdf) | ARX-5 ×2 | Project Aria；双腕 D405；[证据](research/robot_hardware/evidence.md#paper-2607.08436) |
| <a id="paper-2607.08575"></a>2607.08575<br>VLA | [FabriVLA: A Lightweight Vision-Language-Action Model with Conformal Action Chunk Uncertainty](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.08575.pdf) | Unitree D1-T | 立方体入托盘；[证据](research/robot_hardware/evidence.md#paper-2607.08575) |
| <a id="paper-2607.08974"></a>2607.08974<br>VLA | [CLAP: Direct VLM-to-VLA Adaptation via Language-Action Grounding](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.08974.pdf) | UR5e | Robotiq 2F-85；[证据](research/robot_hardware/evidence.md#paper-2607.08974) |
| <a id="paper-2607.09042"></a>2607.09042<br>VLA | [Learning More from Less: Reinforcement Learning from Hindsight](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.09042.pdf) | Franka FR3 | 真实控制；[证据](research/robot_hardware/evidence.md#paper-2607.09042) |
| <a id="paper-2607.09776"></a>2607.09776<br>VLA | [HELP: Human-Efficient Large-Scale Robot Post-Training with Rollout Segmentation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.09776.pdf) | ARX-5 | 自建真实 rollout 数据；后训练；[证据](research/robot_hardware/evidence.md#paper-2607.09776) |
| <a id="paper-2607.09818"></a>2607.09818<br>VLA | [TS-Mask VLA: 2D Temporal-Spatial Masking for Vision-Language-Action Model with Effective Bridging](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.09818.pdf) | UR5e | Robotiq-85；[证据](research/robot_hardware/evidence.md#paper-2607.09818) |
| <a id="paper-2607.10172"></a>2607.10172<br>VLA | [On the Efficiency of LoRA Fine-Tuning for Vision-Language-Action Models in Industrial Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.10172.pdf) | UR5e | Robotiq 2F-85；GELLO 遥操作；[证据](research/robot_hardware/evidence.md#paper-2607.10172) |
| <a id="paper-2607.10180"></a>2607.10180<br>VLA | [ActiveFly-Bench: Aligning Embodied Question Answering with Vision-Language-Action for Aerial Embodied Perception](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.10180.pdf) | 自制 280 mm 四旋翼 | PX4；NUC13；Mid-360；[证据](research/robot_hardware/evidence.md#paper-2607.10180) |
| <a id="paper-2607.11270"></a>2607.11270<br>WAM | [Towards Predictive, Aligned, and Scalable Robot Learning](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.11270.pdf) | Astribot S1 | 22 项任务；[证据](research/robot_hardware/evidence.md#paper-2607.11270) |
| <a id="paper-2607.11498"></a>2607.11498<br>VLA | [See like a Robot: Robot-Centric Pointmaps for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.11498.pdf) | Franka FR3 | 机器人坐标点图；[证据](research/robot_hardware/evidence.md#paper-2607.11498) |
| <a id="paper-2607.12287"></a>2607.12287<br>VLA | [Reducing Temporal Redundancy for Efficient Vision-Language-Action Inference](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.12287.pdf) | Marvin Pro | 双 7-DoF；[证据](research/robot_hardware/evidence.md#paper-2607.12287) |
| <a id="paper-2607.12356"></a>2607.12356<br>VLA | [VistaVLA: Geometry- and Semantic-Aware 3D Gaussian-Grounded VLA for Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.12356.pdf) | AgileX PiPER | 文中 Agile-Piper；[证据](research/robot_hardware/evidence.md#paper-2607.12356) |
| <a id="paper-2607.12571"></a>2607.12571<br>VLA | [TrustVLA: Mechanism-Guided Inference-Time Defense Against Vision-Language-Action Backdoors](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.12571.pdf) | AgileX PiPER-X | 文中 Songling Piper X；[证据](research/robot_hardware/evidence.md#paper-2607.12571) |
| <a id="paper-2607.12659"></a>2607.12659<br>VLA | [Jetson-PI: Towards Onboard Real-Time Robot Control via Foresight-Aligned Asynchronous Inference](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.12659.pdf) | PrimeBot X2-W | 折叠；另列的 Mycobot/Koch/Realman 功耗比较不等同策略闭环；[证据](research/robot_hardware/evidence.md#paper-2607.12659) |
| <a id="paper-2607.12931"></a>2607.12931<br>VLA | [ExToken: Structured Exploration for Efficient Vision-Language-Action Reinforcement Fine-tuning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.12931.pdf) | AgileX Cobot Magic（PiPER 执行臂） | 四臂含两主臂、两从臂；不是四条执行臂；[证据](research/robot_hardware/evidence.md#paper-2607.12931) |
| <a id="paper-2607.12992"></a>2607.12992<br>VLA | [ChunkFlow: Towards Continuity-Consistent Chunked Policy Learning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.12992.pdf) | 型号未明 | 两项真机任务；[证据](research/robot_hardware/evidence.md#paper-2607.12992) |
| <a id="paper-2607.13017"></a>2607.13017<br>WAM | [FlowWAM: Optical Flow as a Unified Action Representation for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.13017.pdf) | Franka FR3；ARX 5 ×2 | 附录E.1；保留原文ARX 5命名；[证据](research/robot_hardware/evidence.md#paper-2607.13017) |
| <a id="paper-2607.13056"></a>2607.13056<br>VLA | [HRIBench: Benchmarking Interaction-Centric Human-Robot Collaboration](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.13056.pdf) | SO-100 | 人机协作；[证据](research/robot_hardware/evidence.md#paper-2607.13056) |
| <a id="paper-2607.13429"></a>2607.13429<br>VLA | [Generalizable VLA Finetuning via Representation Anchoring and Language-Action Alignment](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.13429.pdf) | xArm7 | 30 Hz 策略；100 Hz 底层；[证据](research/robot_hardware/evidence.md#paper-2607.13429) |
| <a id="paper-2607.13597"></a>2607.13597<br>VLA | [Semantic Anchoring for Robotic Action Representations](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.13597.pdf) | AgileX Cobot Magic V2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2607.13597) |
| <a id="paper-2607.13960"></a>2607.13960<br>WAM | [GigaWorld-Policy-0.5: A Faster and Stronger WAM Empowered by AutoResearch](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.13960.pdf) | AgileX PiPER | 单臂；[证据](research/robot_hardware/evidence.md#paper-2607.13960) |
| <a id="paper-2607.14236"></a>2607.14236<br>VLA | [Never Too Late for Force: Accelerating VLA Post-Training with Reactive Force Injection](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.14236.pdf) | Flexiv Rizon 4S | Robotiq 2F-85；力反馈；[证据](research/robot_hardware/evidence.md#paper-2607.14236) |
| <a id="paper-2607.14609"></a>2607.14609<br>VLA | [Representation-Aligned Tactile Grounding for Contact-Rich Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.14609.pdf) | ARX R5 | PaXini 触觉；[证据](research/robot_hardware/evidence.md#paper-2607.14609) |
| <a id="paper-2607.14635"></a>2607.14635<br>VLA | [Action QFormer: Structured Representation Shaping under Action Supervision in Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.14635.pdf) | 导航平台（型号未明） | 真机指令跟随；[证据](research/robot_hardware/evidence.md#paper-2607.14635) |
| <a id="paper-2607.14695"></a>2607.14695<br>VLA | [Reflex: Real-Time VLA Control through Streaming Inference](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.14695.pdf) | AgileX PiPER | 流式推理；[证据](research/robot_hardware/evidence.md#paper-2607.14695) |
| <a id="paper-2607.14698"></a>2607.14698<br>VLA | [Lights, Camera, Malfunction: When Illumination Robustness Leaves VLA Models Blind to Color](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.14698.pdf) | SO-101 Arm Pro | 照明攻击真机实验；[证据](research/robot_hardware/evidence.md#paper-2607.14698) |
| <a id="paper-2607.14852"></a>2607.14852<br>VLA | [Towards Human-like Physical Intelligence: Lifelong Vision-Language-Action Learning for Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.14852.pdf) | xArm（版本未明） | 终身学习五任务流；[证据](research/robot_hardware/evidence.md#paper-2607.14852) |
| <a id="paper-2607.14997"></a>2607.14997<br>WAM | [AeroAct: Action-Centered World-Action Models for Language-Conditioned Quadrotor Flight](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.14997.pdf) | 自制四旋翼（型号未明） | 定性飞行；未报告实飞成功率；[证据](research/robot_hardware/evidence.md#paper-2607.14997) |
| <a id="paper-2607.15330"></a>2607.15330<br>VLA | [Xiaomi-Robotics-1: Scaling Vision-Language-Action Models with over 100K Hours of Real-World Trajectories](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.15330.pdf) | 型号未明 | 不能把 UMI 数据采集器当执行机器人；[证据](research/robot_hardware/evidence.md#paper-2607.15330) |
| <a id="paper-2607.15714"></a>2607.15714<br>VLA | [AC-VLA: Robust Out-of-Distribution Action Execution via Compositional Learning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.15714.pdf) | AgileX PiPER | 6-DoF；[证据](research/robot_hardware/evidence.md#paper-2607.15714) |
| <a id="paper-2607.16636"></a>2607.16636<br>VLA | [PhyAgentOS: A Self-Evolving Operating System for Embodied Agents with Decoupled Cognitive Planning and Physical Execution](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.16636.pdf) | Franka FR3；AgileX PiPER；Dobot Nova2；XLeRobot；Unitree Go2；Unitree G1；Zerith | 系统接入及任务验证；支持列表不全视为实验；[证据](research/robot_hardware/evidence.md#paper-2607.16636) |
| <a id="paper-2607.18016"></a>2607.18016<br>VLA | [Closing the Loop in Humanoid VLA: Persistent 3D Object Tokens for Verifiable Loco-Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.18016.pdf) | Unitree G1 | Dex3-1；[证据](research/robot_hardware/evidence.md#paper-2607.18016) |
| <a id="paper-2607.18231"></a>2607.18231<br>VLA | [FM-VLA: Force-based Memory for Vision-Language-Action Models in Contact-Rich Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.18231.pdf) | AgiBot G1 | 双臂腕部六维力；[证据](research/robot_hardware/evidence.md#paper-2607.18231) |
| <a id="paper-2607.18580"></a>2607.18580<br>VLA | [STeP: Signal Temporal Logic for Precise Specifications for Action Generation with Vision Language Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.18580.pdf) | UR3e | 桌面操作；[证据](research/robot_hardware/evidence.md#paper-2607.18580) |
| <a id="paper-2607.18840"></a>2607.18840<br>WAM | [WorldScape Policy 2.0: Empowering Steerable World Action Modeling with Reasoning-Augmented Memory](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.18840.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2607.18840) |
| <a id="paper-2607.20345"></a>2607.20345<br>VLA | [Closing the Lab-to-Store Gap: A Data-Efficient Post-Training and Experience-Driven Learning VLA Framework for Retail Humanoids](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.20345.pdf) | Unitree G1-Edu | Dex-3；补货；[证据](research/robot_hardware/evidence.md#paper-2607.20345) |
| <a id="paper-2607.21725"></a>2607.21725<br>VLA | [Addressing the Orchestration Gap in Generalist Robots via Physical Agency](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.21725.pdf) | Franka FR3 | DROID 桌面系统；[证据](research/robot_hardware/evidence.md#paper-2607.21725) |
| <a id="paper-2607.22997"></a>2607.22997<br>VLA | [Real2Sim2Real for Vision-Language-Action Manipulation: An AMD ROCm-Based Pipeline](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.22997.pdf) | Franka（版本未明） | Real2Sim2Real；[证据](research/robot_hardware/evidence.md#paper-2607.22997) |
| <a id="paper-2607.22999"></a>2607.22999<br>VLA | [WCM: World-Cognition Model for Generalizable Human-Robot Interaction](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.22999.pdf) | 型号未明 | 九项真实人机交互；[证据](research/robot_hardware/evidence.md#paper-2607.22999) |
| <a id="paper-2607.23783"></a>2607.23783<br>WAM | [$N_0$-TWAM: Scaling Tactile-Native World-Action Model for Contact-Rich Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.23783.pdf) | PiPER 12（双臂）；Flexiv（版本未明） | InTacS1 触觉；[证据](research/robot_hardware/evidence.md#paper-2607.23783) |
| <a id="paper-2607.23784"></a>2607.23784<br>VLA | [A Few Words Go a Long Way: Language Guided Robot Policy Synthesis](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.23784.pdf) | Franka Panda | DROID 配置；Robotiq 2F-85；[证据](research/robot_hardware/evidence.md#paper-2607.23784) |
| <a id="paper-2607.23969"></a>2607.23969<br>WAM | [LeapBot-WA: World-Anchor Action Models via Predictive Latent Alignments](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.23969.pdf) | UR5 | 定性抓放；[证据](research/robot_hardware/evidence.md#paper-2607.23969) |
| <a id="paper-2607.24008"></a>2607.24008<br>VLA | [FutureRTC: Real-Time Robot Execution with Anticipatory-Conditioned Action Chunking](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.24008.pdf) | AgileX Cobot Magic | 双臂实时执行；[证据](research/robot_hardware/evidence.md#paper-2607.24008) |
| <a id="paper-2607.24485"></a>2607.24485<br>VLA | [τ: Learning Touch-Augmented Vision-Language-Action Models from Future Visual Supervision](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.24485.pdf) | Franka FR3 | 触觉增强；[证据](research/robot_hardware/evidence.md#paper-2607.24485) |
| <a id="paper-2607.25516"></a>2607.25516<br>VLA | [A Causality-aware Infer-diagnose-refine Framework for Test-time Modality Adaptation in VLA Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.25516.pdf) | ARX-5 ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2607.25516) |
| <a id="paper-2607.25912"></a>2607.25912<br>VLA | [SAM3D-Guided Object-Centric Representation Alignment for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.25912.pdf) | AgileX PiPER-X | 目标表征对齐；[证据](research/robot_hardware/evidence.md#paper-2607.25912) |
| <a id="paper-2607.25918"></a>2607.25918<br>WAM | [DC-WAM: Dynamic-Centric Visual Supervision and Reasoning for World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.25918.pdf) | AgileX PiPER | 三项任务；[证据](research/robot_hardware/evidence.md#paper-2607.25918) |
| <a id="paper-2607.26513"></a>2607.26513<br>VLA | [Explicit Kinematic Guidance from Analytic Concepts for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.26513.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2607.26513) |
| <a id="paper-2607.26807"></a>2607.26807<br>VLA | [Route by Kinematics, Act by Observation: Kinematics-Supervised Expert Routing in MoE-Augmented VLA](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.26807.pdf) | DIYRobot（自制平台） | 不用仿真本体推断品牌；[证据](research/robot_hardware/evidence.md#paper-2607.26807) |
| <a id="paper-2607.26991"></a>2607.26991<br>VLA | [RL$^2$-VLA: Adaptive RL Latent Compositional Steering with Test-Time Scaling for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.26991.pdf) | AgileX PiPER-X | D405；[证据](research/robot_hardware/evidence.md#paper-2607.26991) |
| <a id="paper-2607.27138"></a>2607.27138<br>VLA | [DLAM: Distributional Latent Actions with Temporal Constraints](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.27138.pdf) | AgileX PiPER | 6-DoF；[证据](research/robot_hardware/evidence.md#paper-2607.27138) |
| <a id="paper-2607.27205"></a>2607.27205<br>VLA | [TurboVLA: Real-Time Vision-Language-Action Model at 32 Hz on an RTX 4090 with <1 GB VRAM](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.27205.pdf) | AgileX PiPER | 四任务；[证据](research/robot_hardware/evidence.md#paper-2607.27205) |
| <a id="paper-2607.27782"></a>2607.27782<br>VLA | [RedFlow: Redirect Failure into Action-Level Corrections for Flow-matching VLA Policy](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.27782.pdf) | AgileX Cobot Magic | 双臂；[证据](research/robot_hardware/evidence.md#paper-2607.27782) |
| <a id="paper-2607.27881"></a>2607.27881<br>VLA | [RoboBRIDGE: A Modular Framework for Bridging Policies to Robust Real-World Robotic Agents](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.27881.pdf) | Franka FR3；UR7e | 原文 Franka Emika Research 3；[证据](research/robot_hardware/evidence.md#paper-2607.27881) |
| <a id="paper-2607.28243"></a>2607.28243<br>WAM | [EgoGenesis: Egocentric World-Action Modeling with Online Anchored Projective Memory and Action-3D RoPE](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.28243.pdf) | 单臂及双臂（型号未明） | 合成视频提升真实任务；[证据](research/robot_hardware/evidence.md#paper-2607.28243) |
| <a id="paper-2607.28391"></a>2607.28391<br>WAM | [TacWAM: Anchor-Guided World Action Model with Mechanics-Aware Tactile Prediction](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.28391.pdf) | AgileX PiPER | Xense G1-WS 是触觉传感器，不是 Unitree G1；[证据](research/robot_hardware/evidence.md#paper-2607.28391) |
| <a id="paper-2607.28405"></a>2607.28405<br>WAM | [QuantWAMs: Calibrating at the Right Granularity for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.28405.pdf) | AgiBot G2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2607.28405) |
| <a id="paper-2607.28993"></a>2607.28993<br>WAM | [ST-WAM: Semantic-Temporal World Action Model for Robust Manipulation under Visual Distribution Shifts](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.28993.pdf) | AgileX PiPER | 6-DoF；[证据](research/robot_hardware/evidence.md#paper-2607.28993) |
| <a id="paper-2607.29235"></a>2607.29235<br>WAM | [FBFM: A Training-Free Asynchronous Feedback Mechanism for Flow-Matching in World-Action Models Execution](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.29235.pdf) | 型号未明 | 真实视频诊断/执行性质需参阅原文；[证据](research/robot_hardware/evidence.md#paper-2607.29235) |
| <a id="paper-2607.29569"></a>2607.29569<br>VLA | [Safe Vision Language Action Models via Barrier Enhanced Flow Matching](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.29569.pdf) | SO-101 | 障碍约束；[证据](research/robot_hardware/evidence.md#paper-2607.29569) |
| <a id="paper-2607.29596"></a>2607.29596<br>VLA | [FibVLA: An Efficient Temporal Vision-Language-Action Model with Fibonacci Sampling](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.29596.pdf) | AgileX PiPER | Pika 遥操作；[证据](research/robot_hardware/evidence.md#paper-2607.29596) |
| <a id="paper-2607.29613"></a>2607.29613<br>VLA | [WCM: A World Critic Model for Vision-Language-Action Reinforcement Learning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2607.29613.pdf) | 型号未明 | 真机强化学习；[证据](research/robot_hardware/evidence.md#paper-2607.29613) |
| <a id="paper-2608.00569"></a>2608.00569<br>VLA | [Latency-Tolerant Cloud-Edge Collaborative Vision-Language-Action Models via Emergent Representational Specialization](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.00569.pdf) | Franka（版本未明） | 小规模 pilot；[证据](research/robot_hardware/evidence.md#paper-2608.00569) |
| <a id="paper-2608.00635"></a>2608.00635<br>WAM | [FlowPilot: Real-Time World-Action Modeling for Agile UAV Navigation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.00635.pdf) | 自制四旋翼（OddityRC 35Pro机架） | 机架不等同整机商业型号；[证据](research/robot_hardware/evidence.md#paper-2608.00635) |
| <a id="paper-2608.00725"></a>2608.00725<br>WAM | [SelfWAM: A Self-Grounded Unified World Action Model for Fast Robot Control](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.00725.pdf) | AgileX ALOHA（臂型号未明） | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.00725) |
| <a id="paper-2608.00793"></a>2608.00793<br>WAM | [DynamicWAM: Dual-Path Motion Conditioning for World-Action Models in Dynamic Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.00793.pdf) | Franka FR3 | CTAG2F90 / Robotiq 2F-140；[证据](research/robot_hardware/evidence.md#paper-2608.00793) |
| <a id="paper-2608.01028"></a>2608.01028<br>VLA | [VLAGuard: A Framework for Evaluating and Mitigating Physical Attention Hijacking in Vision-Language-Action Robots within Wireless Sensor Networks](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.01028.pdf) | AgileX PiPER | 物理传感干扰；[证据](research/robot_hardware/evidence.md#paper-2608.01028) |
| <a id="paper-2608.01066"></a>2608.01066<br>VLA | [OC-VLA++: Monocular Geometry-Guided Cross-View Consistency for Viewpoint-Robust Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.01066.pdf) | Franka Emika Panda | 真机扰动评估；[证据](research/robot_hardware/evidence.md#paper-2608.01066) |
| <a id="paper-2608.01221"></a>2608.01221<br>WAM | [EndoWAM: A Grounded World-Action Model for Generalizable Endoscopic Navigation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.01221.pdf) | 机器人内镜系统（型号未明） | 实体解剖体模；不等同患者临床部署；[证据](research/robot_hardware/evidence.md#paper-2608.01221) |
| <a id="paper-2608.01265"></a>2608.01265<br>VLA | [Hermite Curves as Trajectory Priors for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.01265.pdf) | Franka（版本未明）；Cybopal（单/双臂）；ARX（双臂，版本未明） | Hermite 轨迹先验；[证据](research/robot_hardware/evidence.md#paper-2608.01265) |
| <a id="paper-2608.01381"></a>2608.01381<br>VLA | [DreamTrajectory: Trajectory-Guided Action Generation with World Model Alignment for Mobile Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.01381.pdf) | ARX LIFT（代际未明） | 移动操作；[证据](research/robot_hardware/evidence.md#paper-2608.01381) |
| <a id="paper-2608.01397"></a>2608.01397<br>WAM | [SG-WAM: Self-Guided World Modeling in Geometry-Aware Policy Space](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.01397.pdf) | UR5e | 主视 Azure Kinect；腕 D405；[证据](research/robot_hardware/evidence.md#paper-2608.01397) |
| <a id="paper-2608.01402"></a>2608.01402<br>VLA | [Demystifying When and Why VLAs Fail in Contact-Rich Tasks and How to Fix Them](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.01402.pdf) | Franka FR3 | Bota SensONE；Robotiq；[证据](research/robot_hardware/evidence.md#paper-2608.01402) |
| <a id="paper-2608.01690"></a>2608.01690<br>VLA | [ProtoAct: Turning Wet-Lab Protocols into Embodied Robotic Actions](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.01690.pdf) | 实体湿实验机械臂（型号未明） | 实体三任务；[证据](research/robot_hardware/evidence.md#paper-2608.01690) |
| <a id="paper-2608.01824"></a>2608.01824<br>VLA | [ReTouch: Empowering Contact-Rich Dexterous Manipulation with Online-Refined Tactile Prediction](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.01824.pdf) | UR7e＋XHand | 灵巧触觉；[证据](research/robot_hardware/evidence.md#paper-2608.01824) |
| <a id="paper-2608.01826"></a>2608.01826<br>VLA | [Multi-View Unified Camera Fields: Geometry-Shaped Action-Facing Representations for RGB-Only Multi-Camera VLA Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.01826.pdf) | 型号未明 | 60 次真机 pilot；[证据](research/robot_hardware/evidence.md#paper-2608.01826) |
| <a id="paper-2608.01880"></a>2608.01880<br>WAM | [World Action Models in Real Time: An Empirical Study of Smooth Execution via Asynchronous Deployment](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.01880.pdf) | 双臂平台（型号未明） | 异步平滑执行；[证据](research/robot_hardware/evidence.md#paper-2608.01880) |
| <a id="paper-2608.02197"></a>2608.02197<br>VLA | [Look Where It Matters: Adaptive Visual Refinement for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.02197.pdf) | Franka FR3 | 单外置 D435i；[证据](research/robot_hardware/evidence.md#paper-2608.02197) |
| <a id="paper-2608.02257"></a>2608.02257<br>VLA | [Learning Panorama-Aware VLA for Mobile Manipulation with Whole-Body Teleoperation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.02257.pdf) | AgileX PiPER ×2＋RANGER MINI 3 | 全景移动操作；[证据](research/robot_hardware/evidence.md#paper-2608.02257) |
| <a id="paper-2608.02497"></a>2608.02497<br>VLA | [Grounded Semantic Re-Binding for Robust Instruction Generalization in Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.02497.pdf) | AgileX PiPER ×2 | 14 维双臂动作；[证据](research/robot_hardware/evidence.md#paper-2608.02497) |
| <a id="paper-2608.02580"></a>2608.02580<br>VLA | [Ego2Robot: Scalable Robot Data Synthesis from Egocentric Human Data](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.02580.pdf) | ARX ACone | 双臂各6DoF；[证据](research/robot_hardware/evidence.md#paper-2608.02580) |
| <a id="paper-2608.02958"></a>2608.02958<br>VLA | [ValueFormer: A Causal Transformer Value Function with Stage-Aware Labels for Semi-Autonomous Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.02958.pdf) | Trossen WidowX ×2（版本未明） | 三明治组装；不自动改成 ViperX；[证据](research/robot_hardware/evidence.md#paper-2608.02958) |
| <a id="paper-2608.03231"></a>2608.03231<br>VLA | [Structure-Aware Robust Fine-Tuning: Defending Vision-Language-Action Robots Against Physical Attention Hijacking](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.03231.pdf) | AgileX PiPER | 物理打印补丁；[证据](research/robot_hardware/evidence.md#paper-2608.03231) |
| <a id="paper-2608.03483"></a>2608.03483<br>VLA | [Continue or Replan? Bernoulli-Continuation Policy Learning for Adaptive Horizon Execution](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.03483.pdf) | AgiBot G1 | 继续/重规划；[证据](research/robot_hardware/evidence.md#paper-2608.03483) |
| <a id="paper-2608.03563"></a>2608.03563<br>VLA | [Unified Visuomotor Targets: Supervising VLAs Beyond Physical Actions](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.03563.pdf) | 双臂平台（型号未明） | Lift Pot 等；[证据](research/robot_hardware/evidence.md#paper-2608.03563) |
| <a id="paper-2608.03701"></a>2608.03701<br>WAM | [LiLa-WAM: Lightweight Latent Reasoning World-Action Model for Robotic Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.03701.pdf) | AgileX PiPER | 6-DoF；[证据](research/robot_hardware/evidence.md#paper-2608.03701) |
| <a id="paper-2608.03727"></a>2608.03727<br>VLA | [Track4Action: Distilling World-Centric 3D Tracker into Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.03727.pdf) | AgileX PiPER ×2（ALOHA-style） | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.03727) |
| <a id="paper-2608.04246"></a>2608.04246<br>VLA | [SAFECAST: Robust Failure Detection for VLA Policies with Contrast-Set Training and Calibration](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.04246.pdf) | Franka／DROID（机械臂版本未明） | 真实执行π0／π0-FAST收集rollout并评估失败检测；未证明检测器自身闭环干预；[证据](research/robot_hardware/evidence.md#paper-2608.04246) |
| <a id="paper-2608.04396"></a>2608.04396<br>VLA | [CofactVLA: Deconfounding Vision-Language-Action Models via Counterfactual Intervention](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.04396.pdf) | AgileX PiPER | 6-DoF；[证据](research/robot_hardware/evidence.md#paper-2608.04396) |
| <a id="paper-2608.04404"></a>2608.04404<br>WAM | [Faster-WAM: Efficient Inference-Time Future Conditioning for Robust World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.04404.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.04404) |
| <a id="paper-2608.04633"></a>2608.04633<br>VLA | [Mind-VLA: Instruction-Aware Spatial Representation Alignment for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.04633.pdf) | xArm6 | 五十条演示/任务；[证据](research/robot_hardware/evidence.md#paper-2608.04633) |
| <a id="paper-2608.04657"></a>2608.04657<br>WAM | [MobileWAM: Bridging World Action Models to Mobile Manipulation with Chain-of-Foresight](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.04657.pdf) | ARX LIFT2 | 移动操作；[证据](research/robot_hardware/evidence.md#paper-2608.04657) |
| <a id="paper-2608.04765"></a>2608.04765<br>VLA | [Explicit Language Memory for Long-Horizon Planning in Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.04765.pdf) | 型号未明 | 25 条遥操作；20 次评测；[证据](research/robot_hardware/evidence.md#paper-2608.04765) |
| <a id="paper-2608.04996"></a>2608.04996<br>WAM | [DreamWAM: Beyond RGB Future Prediction for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.04996.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.04996) |
| <a id="paper-2608.05042"></a>2608.05042<br>VLA | [BridgeVLA++: A Data-Efficient, Generalizable, and Memory-Augmented Vision-Language-Action Framework for 3D Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.05042.pdf) | Franka（版本未明）；Dobot CR5A | 跨平台；[证据](research/robot_hardware/evidence.md#paper-2608.05042) |
| <a id="paper-2608.05369"></a>2608.05369<br>VLA | [World-to-Wrist: Task-Conditioned Future Wrist Modeling for Fine-Grained Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.05369.pdf) | AgileX Cobot Magic | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.05369) |
| <a id="paper-2608.05738"></a>2608.05738<br>VLA | [In-Context VLA: Endowing Vision-Language-Action Models with Language via In-Context Post-Training and Agentic Tool Use](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.05738.pdf) | AgiBot G1 | 上下文条件；[证据](research/robot_hardware/evidence.md#paper-2608.05738) |
| <a id="paper-2608.05903"></a>2608.05903<br>WAM | [Robust-WAM: Bridging Generative Pretraining and Semantic Foresight in World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.05903.pdf) | Franka FR3 | 光照泛化；[证据](research/robot_hardware/evidence.md#paper-2608.05903) |
| <a id="paper-2608.05970"></a>2608.05970<br>VLA | [SkillMemo: Expert-guided Skill Memory Framework for Compositional Embodied Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.05970.pdf) | UR5e | WSG-50；[证据](research/robot_hardware/evidence.md#paper-2608.05970) |
| <a id="paper-2608.05999"></a>2608.05999<br>VLA | [Beyond Flat Policies: Hierarchical Post-Training for Embodied Agents in Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.05999.pdf) | 型号未明 | JoyRA-0.1 是模型名；[证据](research/robot_hardware/evidence.md#paper-2608.05999) |
| <a id="paper-2608.06374"></a>2608.06374<br>VLA | [DyPES-VLA: Learning Shared Dynamics Priors and Embodiment-Specific Control for Cross-Embodiment Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.06374.pdf) | Franka FR3；AgileX Cobot Magic；Unitree G1 | Panda / ALOHA-AgileX / Fourier GR-1 为对应仿真；[证据](research/robot_hardware/evidence.md#paper-2608.06374) |
| <a id="paper-2608.06375"></a>2608.06375<br>WAM | [$ω$-0: A Latent Predictive World Action Model for Concurrent Humanoid Loco-Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.06375.pdf) | Unitree G1 | Inspire DexHands；全身移动操作；[证据](research/robot_hardware/evidence.md#paper-2608.06375) |
| <a id="paper-2608.06434"></a>2608.06434<br>VLA | [Fast and Accurate: An Adaptive VLA Inference Framework through Environment-aware Model Selection](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.06434.pdf) | RealMan RM75-6F ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.06434) |
| <a id="paper-2608.06688"></a>2608.06688<br>VLA | [CrossTracer: Cross-Embodiment Navigation via VLA Model Reasoning and Trace Residuals Adapting](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.06688.pdf) | 导航平台（型号未明） | 跨本体导航；[证据](research/robot_hardware/evidence.md#paper-2608.06688) |
| <a id="paper-2608.06729"></a>2608.06729<br>VLA | [AtlasVLA: Persistent World-Ego State Modeling for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.06729.pdf) | Franka（版本未明） | D415；[证据](research/robot_hardware/evidence.md#paper-2608.06729) |
| <a id="paper-2608.06965"></a>2608.06965<br>VLA | [Cross-View Action Consistency for Camera-Robust Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.06965.pdf) | RealMan RM-75 | 15 Hz；[证据](research/robot_hardware/evidence.md#paper-2608.06965) |
| <a id="paper-2608.06994"></a>2608.06994<br>WAM | [Decoupling Intention from Trajectory: A Representational Deduction Framework for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.06994.pdf) | AgiBot G1 | 双臂人形；[证据](research/robot_hardware/evidence.md#paper-2608.06994) |
| <a id="paper-2608.07314"></a>2608.07314<br>VLA | [TEMPO: Semantic-Action Decoupled RL Post-Training for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.07314.pdf) | Synria Alicia-D | 6-DoF＋夹爪；[证据](research/robot_hardware/evidence.md#paper-2608.07314) |
| <a id="paper-2608.07596"></a>2608.07596<br>VLA | [LIRA: Local Cross-Layer Information Routing for Vision-Language-Action Decoding](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.07596.pdf) | Franka FR3 | Franka Hand；[证据](research/robot_hardware/evidence.md#paper-2608.07596) |
| <a id="paper-2608.07619"></a>2608.07619<br>VLA | [GWM-VLA: Geometry-Aware Latent World Modeling for Vision-Language-Action Learning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.07619.pdf) | SO-101 | 几何潜在世界建模；[证据](research/robot_hardware/evidence.md#paper-2608.07619) |
| <a id="paper-2608.08023"></a>2608.08023<br>WAM | [4D-WAM: Infusing Spatiotemporal Awareness into World Action Models through Trajectory Fields](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.08023.pdf) | ARX LIFT2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.08023) |
| <a id="paper-2608.08558"></a>2608.08558<br>WAM | [Vid2WAM: Distilling Video Diffusion Priors into World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.08558.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.08558) |
| <a id="paper-2608.08839"></a>2608.08839<br>WAM | [SG-WAM: Text-Grounded and Spatial-aware Semantic Guidance for World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.08839.pdf) | AgileX Cobot（版本未明） | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.08839) |
| <a id="paper-2608.09125"></a>2608.09125<br>VLA | [Trajectory Divergence Horizon Decision for Reliable Dual-Arm Surgical Subtask Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.09125.pdf) | RealMan RM65-B ×2 | dVRK 风格末端，不是 da Vinci 整机；[证据](research/robot_hardware/evidence.md#paper-2608.09125) |
| <a id="paper-2608.09381"></a>2608.09381<br>WAM/VLA | [JEPA-WAM: Learning Vision-Language-Action Policies with Joint-Embedding World Modeling](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.09381.pdf) | AgileX Cobot Magic | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.09381) |
| <a id="paper-2608.09410"></a>2608.09410<br>VLA | [Skills in Weights, Memory in Code: Hybrid Learning for Memory-Dependent Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.09410.pdf) | SO-101 | 记忆任务；[证据](research/robot_hardware/evidence.md#paper-2608.09410) |
| <a id="paper-2608.09492"></a>2608.09492<br>WAM | [Rethink Before You Execute: Adaptive Execution for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.09492.pdf) | 双臂平台（型号未明） | 三项真实任务；[证据](research/robot_hardware/evidence.md#paper-2608.09492) |
| <a id="paper-2608.09730"></a>2608.09730<br>WAM/VLA | [World Tokens: Enhancing Embodied Policies with Training-Time World Modeling](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.09730.pdf) | Galaxea R1 Pro | World Tokens；[证据](research/robot_hardware/evidence.md#paper-2608.09730) |
| <a id="paper-2608.09771"></a>2608.09771<br>VLA | [SLIM-0.5B: Learning Action-Grounded Predictive Latents for Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.09771.pdf) | 型号未明 | 五项真实任务；[证据](research/robot_hardware/evidence.md#paper-2608.09771) |
| <a id="paper-2608.10232"></a>2608.10232<br>WAM | [FACT: Failure-Aware Causal Training for World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.10232.pdf) | I2RT YAM ×2 | D435主视角与两台D405腕相机；GELLO遥操作；[证据](research/robot_hardware/evidence.md#paper-2608.10232) |
| <a id="paper-2608.10393"></a>2608.10393<br>VLA | [Hidden in Plain Sight: Diffusion-Based Unrestricted Robotic Attacks on Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.10393.pdf) | Franka（版本未明） | 7-DoF；[证据](research/robot_hardware/evidence.md#paper-2608.10393) |
| <a id="paper-2608.10718"></a>2608.10718<br>VLA | [TCAM for Autonomous Deformable Manipulation: The RMC2 Champion System for WBCD 2026 Track 4](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.10718.pdf) | ARX X5 | 可变形物体操作；[证据](research/robot_hardware/evidence.md#paper-2608.10718) |
| <a id="paper-2608.10756"></a>2608.10756<br>VLA | [Embodied Multimodal Grounding for Open-Vocabulary Mobile Manipulation via Semantic 3D Gaussian Splatting](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.10756.pdf) | Unitree Go2 Edu＋Synria Alicia-D | 移动底盘＋操作臂；[证据](research/robot_hardware/evidence.md#paper-2608.10756) |
| <a id="paper-2608.10780"></a>2608.10780<br>WAM | [StageWAM: Joint-Embedding Stage Prediction for World-Action Models in Robot Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.10780.pdf) | ARX LIFT2（ARX R5） | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.10780) |
| <a id="paper-2608.10860"></a>2608.10860<br>WAM | [Flex-$π$: A Multi-Stream World-Action Model with Compute Flexibility](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.10860.pdf) | I2RT YAM ×2 | 双臂共14维命令含夹爪；[证据](research/robot_hardware/evidence.md#paper-2608.10860) |
| <a id="paper-2608.11671"></a>2608.11671<br>VLA | [StellaVLA: In-Context Structured Demonstration for Generalizable Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.11671.pdf) | AgileX PiPER | 示例条件泛化；[证据](research/robot_hardware/evidence.md#paper-2608.11671) |
| <a id="paper-2608.11739"></a>2608.11739<br>VLA | [G0.5: One Autoregressive Stream for Robot Reasoning and Action](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.11739.pdf) | Franka FR3 | 另有六设定多本体，见补充证据；[证据](research/robot_hardware/evidence.md#paper-2608.11739) |
| <a id="paper-2608.12416"></a>2608.12416<br>VLA | [RoboSynChallenge: Mastering Real-World Dexterity via Generalizing Synthesized Manipulation Skills](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.12416.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.12416) |
| <a id="paper-2608.13924"></a>2608.13924<br>VLA | [BICPO-VLA: Behavior-Identified Continuation Preference Optimization for Smooth Asynchronous Vision-Language-Action Control](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.13924.pdf) | 型号未明 | 六项真机任务；[证据](research/robot_hardware/evidence.md#paper-2608.13924) |
| <a id="paper-2608.14024"></a>2608.14024<br>VLA | [SSP: An Event-Matched Syn2Sim2Phy Cross-Domain Evaluation Framework for Autonomous Driving VLA Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.14024.pdf) | 实车测试平台（车型未明） | 封闭场地；不是 NAVSIM-only；[证据](research/robot_hardware/evidence.md#paper-2608.14024) |
| <a id="paper-2608.14047"></a>2608.14047<br>VLA | [Evolve Vision-Language-Action Model into an Agent with On-the-fly Tool-use](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.14047.pdf) | Astribot S1 | 16 DoF；[证据](research/robot_hardware/evidence.md#paper-2608.14047) |
| <a id="paper-2608.14379"></a>2608.14379<br>VLA | [Reflex: Enabling Fast and Predictive Vision-Language-Action Models for Reaction-Critical Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.14379.pdf) | AgileX PiPER | 预测式快速执行；[证据](research/robot_hardware/evidence.md#paper-2608.14379) |
| <a id="paper-2608.14822"></a>2608.14822<br>VLA | [Imagining Recovery: Inference-Time Counterfactual Realignment for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.14822.pdf) | xArm6 | 世界模型推理纠正；[证据](research/robot_hardware/evidence.md#paper-2608.14822) |
| <a id="paper-2608.15009"></a>2608.15009<br>VLA | [ForceU-VLA: A Force-Aware Vision-Language-Action Model for Embodied Ultrasound Scanning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.15009.pdf) | UR7e | HISENSE HD60；FT300-S；人体超声；[证据](research/robot_hardware/evidence.md#paper-2608.15009) |
| <a id="paper-2608.15026"></a>2608.15026<br>VLA | [PACE: Phase-Progress-Aware Credit for Long-Horizon Embodied Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.15026.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.15026) |
| <a id="paper-2608.15139"></a>2608.15139<br>VLA | [StructRL: Structured Action-Space Exploration for Flow-Based VLAs](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.15139.pdf) | Franka FR3 | 接触插入与抓取；[证据](research/robot_hardware/evidence.md#paper-2608.15139) |
| <a id="paper-2608.15269"></a>2608.15269<br>VLA | [Remember Smarter: Visual History Compressor and Hyperbolic Experience Space for Robotic Memory](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.15269.pdf) | 型号未明 | 长程记忆；[证据](research/robot_hardware/evidence.md#paper-2608.15269) |
| <a id="paper-2608.15285"></a>2608.15285<br>VLA | [PhaseLoRA: Control-Regime-Conditioned Low-Rank Adaptation for Continuous-Action Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.15285.pdf) | AgileX PiPER | 相位适配；[证据](research/robot_hardware/evidence.md#paper-2608.15285) |
| <a id="paper-2608.15475"></a>2608.15475<br>VLA | [Bit-Flip Attacks on Vision-Language-Action Models: Action-Decoding Architecture Shapes the Vulnerability](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.15475.pdf) | 6-DoF 机械臂（型号未明） | 物理抓放；位翻转不是硬件物理故障注入；[证据](research/robot_hardware/evidence.md#paper-2608.15475) |
| <a id="paper-2608.15680"></a>2608.15680<br>VLA | [Robo-Dopamine 2.0: History-Conditioned and OOD-Aware Process Reward Modeling for Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.15680.pdf) | Franka ×2（版本未明） | Insert Square；[证据](research/robot_hardware/evidence.md#paper-2608.15680) |
| <a id="paper-2608.15816"></a>2608.15816<br>VLA | [ViTaR: Visuo-Tactile Residual Adaptation for Foundation VLA Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.15816.pdf) | RealMan RM-65B | DH 夹爪；9DTact；[证据](research/robot_hardware/evidence.md#paper-2608.15816) |
| <a id="paper-2608.15875"></a>2608.15875<br>VLA | [GigaBrain-0.7: Scaling Embodied Foundation Models to Emergent Capabilities with a Three-System Architecture](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.15875.pdf) | AgileX PiPER；AgileX PiPER-X；Maker H01 | 多本体；[证据](research/robot_hardware/evidence.md#paper-2608.15875) |
| <a id="paper-2608.16074"></a>2608.16074<br>VLA | [US-VLA: An Ultrasound Vision-Language-Action Model for Embodied Abdomina](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.16074.pdf) | UR7e | 人体超声；[证据](research/robot_hardware/evidence.md#paper-2608.16074) |
| <a id="paper-2608.16172"></a>2608.16172<br>VLA | [SparkVLA: Stop-Aware Hierarchical VLA with Adaptive Action Chunking for Long-Horizon Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.16172.pdf) | AIRBOT Play ×2 | 双臂真实控制；[证据](research/robot_hardware/evidence.md#paper-2608.16172) |
| <a id="paper-2608.16476"></a>2608.16476<br>VLA | [Exposing the Long-tail in Embodied Urban Navigation via Scalable Learning from In-the-Wild Videos](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.16476.pdf) | 移动机器人（型号未明） | 城市真实导航定性展示；[证据](research/robot_hardware/evidence.md#paper-2608.16476) |
| <a id="paper-2608.16503"></a>2608.16503<br>VLA | [NebulaVLA: A Dual-Frequency Vision-Language-Action Model With Guide Action for Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.16503.pdf) | AgiBot A2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.16503) |
| <a id="paper-2608.16837"></a>2608.16837<br>VLA | [HAF: Adapting Generalist VLAs to Humanoid Whole-Body Loco-manipulation via Hierarchical Action Flow and Spectral Latent RL](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.16837.pdf) | TienKung 2.0；TienKung 3.0 | 两代人形平台；[证据](research/robot_hardware/evidence.md#paper-2608.16837) |
| <a id="paper-2608.16885"></a>2608.16885<br>VLA | [$τ_0$-VLA: a Hierarchical Robot Foundation Model with World-Model-Guided Test-Time Computation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.16885.pdf) | AgiBot G1；ARX AC One；Franka FR3 ×2 | 不同真实任务分别部署；[证据](research/robot_hardware/evidence.md#paper-2608.16885) |
| <a id="paper-2608.17209"></a>2608.17209<br>WAM/VLA | [Teach and Grow: An Agent-Centered Architecture for General Robot Learning](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.17209.pdf) | 型号未明 | 机器人技能架构；[证据](research/robot_hardware/evidence.md#paper-2608.17209) |
| <a id="paper-2608.17423"></a>2608.17423<br>VLA | [Prism-GRPO: Faster VLA Policy Optimization via Splitting Same-outcome Groups](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.17423.pdf) | AgileX PiPER | 强化学习；[证据](research/robot_hardware/evidence.md#paper-2608.17423) |
| <a id="paper-2608.17453"></a>2608.17453<br>VLA | [EATR-Stereo: Embodiment-Aware Token Routing of Paired Stereo Evidence for Humanoid Vision-Language-Action Control](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.17453.pdf) | Omega | 人形平台；具体厂家/代际不补猜；[证据](research/robot_hardware/evidence.md#paper-2608.17453) |
| <a id="paper-2608.17600"></a>2608.17600<br>VLA | [LIBERO-VIFO: Benchmarking the Capability and Safety of Visual Cue Following in Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.17600.pdf) | AgileX PiPER | 叠杯；[证据](research/robot_hardware/evidence.md#paper-2608.17600) |
| <a id="paper-2608.17717"></a>2608.17717<br>VLA | [CompCPZ: Preserving Multi-Modal Intent in Language-Guided Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.17717.pdf) | Unitree Go2 | 真机导航；Franka 操作为仿真；[证据](research/robot_hardware/evidence.md#paper-2608.17717) |
| <a id="paper-2608.18077"></a>2608.18077<br>WAM | [Hydra-0: Action Flow for Generalist World Modeling and Control](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.18077.pdf) | 机器人型号未明 | Figure10柔性管弯曲真实执行；其他部分含离线世界预测；[证据](research/robot_hardware/evidence.md#paper-2608.18077) |
| <a id="paper-2608.19490"></a>2608.19490<br>VLA | [Fine-Tuning VLAs with Self-Demonstrated Generative Control for Multi-Task Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.19490.pdf) | ALOHA / Interbotix（臂型号未明） | 原文 Fig.10；[证据](research/robot_hardware/evidence.md#paper-2608.19490) |
| <a id="paper-2608.19574"></a>2608.19574<br>WAM | [HiTac-WAM: A Hierarchical Tactile World Action Model for Contact-Rich Robot Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.19574.pdf) | IMETA-Y1 | DM-TacW2触觉；[证据](research/robot_hardware/evidence.md#paper-2608.19574) |
| <a id="paper-2608.19589"></a>2608.19589<br>VLA | [OrthoSkillVLA: Continual Skill Learning via Gradient-Informed Skill Subspace Adaptation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.19589.pdf) | xArm7＋Inspire 灵巧手 | 持续技能学习；[证据](research/robot_hardware/evidence.md#paper-2608.19589) |
| <a id="paper-2608.20114"></a>2608.20114<br>WAM | [DECOWAM: Decoupled Whole-Body World-Action Model for Legged Mobile Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.20114.pdf) | 足式移动操作平台（型号未明） | 79 次/方法；[证据](research/robot_hardware/evidence.md#paper-2608.20114) |
| <a id="paper-2608.20478"></a>2608.20478<br>VLA | [EndoLIFT: Language-Disambiguated Latent-Conditioned Rectified Flow for Bidirectional Endoscopic Control](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.20478.pdf) | 自制机器人内镜系统 | 进给＋双轴弯曲；实体体模；[证据](research/robot_hardware/evidence.md#paper-2608.20478) |
| <a id="paper-2608.20735"></a>2608.20735<br>WAM/VLA | [ForeTime-VLA: Causal Future-Token Distillation from a World Action Model for Conveyor-Belt Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.20735.pdf) | 机械臂＋传送带（型号未明） | 动态抓取；[证据](research/robot_hardware/evidence.md#paper-2608.20735) |
| <a id="paper-2608.20791"></a>2608.20791<br>VLA | [CertVLA: Certified Defense against Physical Visual Attacks for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.20791.pdf) | AgileX PiPER | 物理攻击防御；[证据](research/robot_hardware/evidence.md#paper-2608.20791) |
| <a id="paper-2608.21740"></a>2608.21740<br>VLA | [CounterAlign: Counterfactual Supervision for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.21740.pdf) | TX-G2（兼容 AgiBot G2） | 保留原文命名，不直接归为 AgiBot 出厂型号；[证据](research/robot_hardware/evidence.md#paper-2608.21740) |
| <a id="paper-2608.22067"></a>2608.22067<br>WAM | [Inferring Action from Future Latent State for Robotic Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.22067.pdf) | Astribot S1 | 四任务；[证据](research/robot_hardware/evidence.md#paper-2608.22067) |
| <a id="paper-2608.22403"></a>2608.22403<br>WAM | [LD4WAM: Learning Latent Dynamics from Human Videos for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.22403.pdf) | AgileX PiPER ×2＋Pika夹爪；Tianji 双臂＋Wuji手 | 平行夹爪与灵巧手；[证据](research/robot_hardware/evidence.md#paper-2608.22403) |
| <a id="paper-2608.22419"></a>2608.22419<br>VLA | [Robust Bimanual Vision-Language-Action Models via Embarrassingly Simple Modality Masking](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.22419.pdf) | AgileX Cobot（版本未明） | 双臂长程；[证据](research/robot_hardware/evidence.md#paper-2608.22419) |
| <a id="paper-2608.22800"></a>2608.22800<br>VLA | [Triplet2Track: A Hierarchical System with Object-Centric Representations for Reliable Long-Horizon Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.22800.pdf) | Franka（版本未明） | 7-DoF；[证据](research/robot_hardware/evidence.md#paper-2608.22800) |
| <a id="paper-2608.22869"></a>2608.22869<br>VLA | [UniMem: Unifying Multimodal Memory and Control for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.22869.pdf) | xArm6 | Franka 是仿真；[证据](research/robot_hardware/evidence.md#paper-2608.22869) |
| <a id="paper-2608.22990"></a>2608.22990<br>VLA | [InstructMove: A Text-Indispensable Benchmark for Instruction-Following Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.22990.pdf) | AgileX PiPER-X ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.22990) |
| <a id="paper-2608.23138"></a>2608.23138<br>VLA | [Pointing-VLA: Typed Spatial Grounding Interfaces for Vision-Language-Action Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.23138.pdf) | AgileX PiPER | 指定部位抓放；[证据](research/robot_hardware/evidence.md#paper-2608.23138) |
| <a id="paper-2608.23224"></a>2608.23224<br>VLA | [Think Only When Needed: Prompt-Authority Control for Selective Slow-Path Intervention in Vision-Language-Action Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.23224.pdf) | AgileX PiPER | 150 次/方法；[证据](research/robot_hardware/evidence.md#paper-2608.23224) |
| <a id="paper-2608.23320"></a>2608.23320<br>VLA | [ROS2SmolVLA: Enabling Small Vision-Language-Action Models for Integration into Industrial-Grade Lightweight Robots](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.23320.pdf) | UR10e | Robotiq Hand-E；[证据](research/robot_hardware/evidence.md#paper-2608.23320) |
| <a id="paper-2608.23478"></a>2608.23478<br>VLA | [Act with Intent: Distilling Behavior Intent for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.23478.pdf) | SO-101 ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.23478) |
| <a id="paper-2608.24042"></a>2608.24042<br>VLA | [Hierarchical Skill Retrieval for Data-Efficient Adaptation of Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.24042.pdf) | xArm（版本未明） | DROID 是源数据；[证据](research/robot_hardware/evidence.md#paper-2608.24042) |
| <a id="paper-2608.24101"></a>2608.24101<br>VLA | [TrAct: Bridging Robot Control and Visual Prediction with Visual Tracks](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.24101.pdf) | Franka Panda | 400 条真实示范；[证据](research/robot_hardware/evidence.md#paper-2608.24101) |
| <a id="paper-2608.24115"></a>2608.24115<br>VLA | [PonderPounce: A Pretrained MLLM as an Episode Context Engine for Robot Control](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.24115.pdf) | Trossen ALOHA Stationary AI Kit（右臂） | 原文附录E含59次真机试验；阅读报告漏记；[证据](research/robot_hardware/evidence.md#paper-2608.24115) |
| <a id="paper-2608.24603"></a>2608.24603<br>VLA | [Gripper-aware Vision Language Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.24603.pdf) | UR5 | Robotiq 2F-85；[证据](research/robot_hardware/evidence.md#paper-2608.24603) |
| <a id="paper-2608.24714"></a>2608.24714<br>WAM | [GaussianWAM: Distilling Geometry and Semantics from 3D Gaussian Fields into World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.24714.pdf) | UR7e ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.24714) |
| <a id="paper-2608.24882"></a>2608.24882<br>WAM | [Latent Action as Intention Enables Efficient Future Imagination for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.24882.pdf) | xArm7 | 潜在意图与未来想象；[证据](research/robot_hardware/evidence.md#paper-2608.24882) |
| <a id="paper-2608.24959"></a>2608.24959<br>VLA | [GaussVLA: Geometry-Aware Spatial Reasoning for Vision-Language-Action Model](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.24959.pdf) | SO-101 | 6-DoF；[证据](research/robot_hardware/evidence.md#paper-2608.24959) |
| <a id="paper-2608.25308"></a>2608.25308<br>VLA | [V-Link: Recovering Lost Visual Representations in Action DiT for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.25308.pdf) | AgiBot A3 Ultra | 50 次/任务；[证据](research/robot_hardware/evidence.md#paper-2608.25308) |
| <a id="paper-2608.25395"></a>2608.25395<br>VLA | [A Taxonomy of Construction Task Activities for Robot Workers](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.25395.pdf) | Dobot CR3＋CRAFT 手 | 脚本驱动真机演示；不是 VLA 自主闭环证据；[证据](research/robot_hardware/evidence.md#paper-2608.25395) |
| <a id="paper-2608.25585"></a>2608.25585<br>VLA | [RA-VLA: Retrieval-Augmented VLA for Test-Time Adaptation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.25585.pdf) | UR5e | Robotiq 2F-85；[证据](research/robot_hardware/evidence.md#paper-2608.25585) |
| <a id="paper-2608.25659"></a>2608.25659<br>VLA | [GaussianDream++: Efficient 3D Gaussian World Modeling for Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.25659.pdf) | 型号未明 | 两任务、布局及语义扰动；[证据](research/robot_hardware/evidence.md#paper-2608.25659) |
| <a id="paper-2608.25757"></a>2608.25757<br>VLA | [LM-X: Explainable Action Modeling with Progress, Event, and Uncertainty Prediction for Generalist Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.25757.pdf) | AgileX ALOHA；Astribot S1；Tianji Marvin；Loong S1 | 四种本体；[证据](research/robot_hardware/evidence.md#paper-2608.25757) |
| <a id="paper-2608.25798"></a>2608.25798<br>VLA | [TacForcing: Streaming Action Generation with Execution-Time Tactile Feedback](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.25798.pdf) | RealMan RM75 ×2＋Sharpa Wave ×2 | 双臂灵巧触觉；[证据](research/robot_hardware/evidence.md#paper-2608.25798) |
| <a id="paper-2608.25864"></a>2608.25864<br>VLA | [MA-VLA: Multi-Arm Vision-Language-Action Model for Collaboration and Compositional Generalization](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.25864.pdf) | SO-101 | 多臂实验；[证据](research/robot_hardware/evidence.md#paper-2608.25864) |
| <a id="paper-2608.26058"></a>2608.26058<br>VLA | [One Policy, Many Embodiments: Unified Camera-Centric Action Geometry Pre-training for Heterogeneous Embodied Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.26058.pdf) | AgileX PiPER | 相机坐标动作；[证据](research/robot_hardware/evidence.md#paper-2608.26058) |
| <a id="paper-2608.26067"></a>2608.26067<br>VLA | [StreamPI: Streaming Multimodal Temporal Modeling for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.26067.pdf) | AgileX PiPER-X | ALOHA-style；[证据](research/robot_hardware/evidence.md#paper-2608.26067) |
| <a id="paper-2608.26103"></a>2608.26103<br>WAM | [Zero-WAM: In-Context World-Action Modeling from Human Videos for Open-Ended Task Generalization](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.26103.pdf) | Franka ×2（版本未明） | 人机上下文示例；[证据](research/robot_hardware/evidence.md#paper-2608.26103) |
| <a id="paper-2608.26578"></a>2608.26578<br>VLA | [TrapVLA: Trapping Vision-Language-Action Models in Configured Failure Modes](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.26578.pdf) | ROKAE 6DoF（版本未明） | 对抗攻击真实验证；[证据](research/robot_hardware/evidence.md#paper-2608.26578) |
| <a id="paper-2608.26645"></a>2608.26645<br>VLA | [FLARE: A Failure-Aware Framework for Autonomous Correction and Recovery in Visual-Language Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.26645.pdf) | AgileX PiPER | 失败恢复；[证据](research/robot_hardware/evidence.md#paper-2608.26645) |
| <a id="paper-2608.26821"></a>2608.26821<br>VLA | [TemporalFlow-VLA: Learning Physically Grounded Execution History for Long-Horizon Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.26821.pdf) | AgiBot A3 | 未补 Ultra 后缀；[证据](research/robot_hardware/evidence.md#paper-2608.26821) |
| <a id="paper-2608.27033"></a>2608.27033<br>WAM | [Riemann-1.0: An Embodied World Action Model for Physical AI](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.27033.pdf) | Tianji Marvin | 双臂；[证据](research/robot_hardware/evidence.md#paper-2608.27033) |
| <a id="paper-2608.27079"></a>2608.27079<br>VLA | [GRAFT: Grounded and Efficient Online Reinforcement Adaptation for Fine-Grained Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.27079.pdf) | JAKA（版本未明） | 精细湿实验操作；[证据](research/robot_hardware/evidence.md#paper-2608.27079) |
| <a id="paper-2608.27384"></a>2608.27384<br>VLA | [FlashVLA: Streaming Action Decoding for Fast and Asynchronous VLA Inference](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.27384.pdf) | Franka（版本未明） | 流式动作；[证据](research/robot_hardware/evidence.md#paper-2608.27384) |
| <a id="paper-2608.27550"></a>2608.27550<br>WAM/VLA | [Beyond Data Scaling: Representation-Centric Continued Pre-training for Vision-Language-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.27550.pdf) | Franka FR3（单/双臂） | 表征续预训练；[证据](research/robot_hardware/evidence.md#paper-2608.27550) |
| <a id="paper-2608.27609"></a>2608.27609<br>VLA | [PHR-VLA: Planning Horizon Reasoning for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.27609.pdf) | Franka（版本未明） | 拆卸；[证据](research/robot_hardware/evidence.md#paper-2608.27609) |
| <a id="paper-2608.28108"></a>2608.28108<br>VLA | [DeicticVLA: Unifying Instruction Modes Based on Language and Deictic Gestures in a Single VLA](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.28108.pdf) | UR5e | Robotiq 2F-85；UMI 式手指；[证据](research/robot_hardware/evidence.md#paper-2608.28108) |
| <a id="paper-2608.28995"></a>2608.28995<br>WAM | [Hydra: A Navigation World Action Model with Discrete Latent Planning and Continuous Flow-Matching Execution](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.28995.pdf) | Boston Dynamics Spot；Clearpath Jackal | 真机导航；[证据](research/robot_hardware/evidence.md#paper-2608.28995) |
| <a id="paper-2608.29078"></a>2608.29078<br>VLA | [DREAM: Deployment-Time Demonstration Generation via Real-to-Sim for Scalable Policy Adaptation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.29078.pdf) | xArm7 | 部署期合成演示；[证据](research/robot_hardware/evidence.md#paper-2608.29078) |
| <a id="paper-2608.29208"></a>2608.29208<br>VLA | [AdaVLA: Adaptive Step Flow Matching for Training-free Acceleration of Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.29208.pdf) | SO-ARM101 | 即 SO-101 命名变体；[证据](research/robot_hardware/evidence.md#paper-2608.29208) |
| <a id="paper-2608.29432"></a>2608.29432<br>VLA | [SMILE: Smooth Motion for Improved Long-Horizon VLA Execution](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.29432.pdf) | UFACTORY xArm7 | 真实操作；[证据](research/robot_hardware/evidence.md#paper-2608.29432) |
| <a id="paper-2608.29537"></a>2608.29537<br>VLA | [AGM: Achievement-Grounded Memory for Closed-Loop Agents with Frozen VLA Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.29537.pdf) | AgileX PiPER | 6-DoF＋夹爪；[证据](research/robot_hardware/evidence.md#paper-2608.29537) |
| <a id="paper-2608.29749"></a>2608.29749<br>VLA | [DriftingVLA: Native One-Step Vision-Language-Action Generation via Per-Dimension Temporal Drifting](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.29749.pdf) | UR5（单臂及双臂） | 六任务；[证据](research/robot_hardware/evidence.md#paper-2608.29749) |
| <a id="paper-2608.29967"></a>2608.29967<br>VLA | [Training-Free Action Correction for VLA Model Failures via Language Feedback](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.29967.pdf) | xArm7 | Robotiq 2F-85；[证据](research/robot_hardware/evidence.md#paper-2608.29967) |
| <a id="paper-2608.30237"></a>2608.30237<br>WAM | [Motus2: A Self-Evolving General World Model for Dexterous Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.30237.pdf) | Tianji 机械臂/平台（版本未明）；Wuji 手；Sharpa 手 | 臂与手分开记录；[证据](research/robot_hardware/evidence.md#paper-2608.30237) |
| <a id="paper-2608.30289"></a>2608.30289<br>VLA | [CometVLA: Co-Training on an Embodied Data Pyramid towards Physical Understanding](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.30289.pdf) | G1（厂商待核） | 不能凭 G1 推断 Unitree/AgiBot/Galbot；[证据](research/robot_hardware/evidence.md#paper-2608.30289) |
| <a id="paper-2608.30643"></a>2608.30643<br>VLA | [Temporal Forcing: 4D Representation Alignment for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.30643.pdf) | UR3 | 真实操作；[证据](research/robot_hardware/evidence.md#paper-2608.30643) |
| <a id="paper-2609.00188"></a>2609.00188<br>WAM | [ZimaBlue: Evolving Generalizable World Action Models through Scalable Video Pre-training](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.00188.pdf) | Franka 7DoF（版本未明） | 真实控制；[证据](research/robot_hardware/evidence.md#paper-2609.00188) |
| <a id="paper-2609.00908"></a>2609.00908<br>VLA | [Knowing When to Stop: Adaptive Action Chunking via Internal Cross-Attention Dynamics in VLAs](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.00908.pdf) | ARX R5（从臂）；ARX X5（主臂） | 不把主臂算策略执行臂；[证据](research/robot_hardware/evidence.md#paper-2609.00908) |
| <a id="paper-2609.02159"></a>2609.02159<br>WAM | [World-Coherent Decoding: Self-Verifying Test-Time Planning for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.02159.pdf) | Franka（版本未明） | 敲击；[证据](research/robot_hardware/evidence.md#paper-2609.02159) |
| <a id="paper-2609.02531"></a>2609.02531<br>WAM | [Spatially Aware World Action Model via Geometric Latent Diffusion](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.02531.pdf) | UR5 | 空间几何泛化；[证据](research/robot_hardware/evidence.md#paper-2609.02531) |
| <a id="paper-2609.02546"></a>2609.02546<br>VLA | [ZETA: A Controlled Study of Zero-Shot Cross-Embodiment VLA Transfer for Tabletop Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.02546.pdf) | Franka（版本未明）；UR（版本未明）；人形机器人（型号未明）；PiPER＋四足底盘（型号未明） | 含Robotiq和UMI末端配置；附表24区分8个配置行，不等同8种机械臂；[证据](research/robot_hardware/evidence.md#paper-2609.02546) |
| <a id="paper-2609.02653"></a>2609.02653<br>VLA | [HINT: Human-Intent Inception for Long-Horizon Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.02653.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.02653) |
| <a id="paper-2609.03142"></a>2609.03142<br>VLA | [Sensing Which Modality Matters: Evidence-Gated Regularization for Robust VLA Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.03142.pdf) | Kinova ×2（版本未明）；Mitsubishi MELFA ASSISTA | ASSISTA 双 GelSight；[证据](research/robot_hardware/evidence.md#paper-2609.03142) |
| <a id="paper-2609.03276"></a>2609.03276<br>VLA | [R2S-Eval: Robot Evaluation with Real-to-Sim Calibration via Vision-Language Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.03276.pdf) | SharpaNorth＋SharpaWave | 真实/仿真相关性评测；[证据](research/robot_hardware/evidence.md#paper-2609.03276) |
| <a id="paper-2609.03483"></a>2609.03483<br>VLA | [Air-Ground Collaborative Vision-and-Language Navigation via Shared Bird's-Eye Maps](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.03483.pdf) | 自制四旋翼＋全向地面车（型号未明） | 空地协同导航；[证据](research/robot_hardware/evidence.md#paper-2609.03483) |
| <a id="paper-2609.03591"></a>2609.03591<br>VLA | [Scaling Bimanual Household Manipulation from 1,500 hours of Demonstrations to On-Policy Corrections](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.03591.pdf) | 双臂平台（型号未明） | 真实示范与在策略纠正；[证据](research/robot_hardware/evidence.md#paper-2609.03591) |
| <a id="paper-2609.03681"></a>2609.03681<br>VLA | [WISE: World-model-guided Imagination Scheduling for Efficient Post-training of Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.03681.pdf) | Galaxea R1 Lite | 四任务；[证据](research/robot_hardware/evidence.md#paper-2609.03681) |
| <a id="paper-2609.03889"></a>2609.03889<br>VLA | [FWBC-VLA: Force-Aware Whole-Body Compensation for Contact-Rich Loco-Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.03889.pdf) | DeepRobotics M20S＋CM1 | 轮足底盘＋机械臂；[证据](research/robot_hardware/evidence.md#paper-2609.03889) |
| <a id="paper-2609.04193"></a>2609.04193<br>WAM/VLA | [GIFT: Guided Intermediate Feature Training via Action-Oriented Structural Supervision for Robotic Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.04193.pdf) | xArm7；ARX X5 ×2 | 单/双臂；[证据](research/robot_hardware/evidence.md#paper-2609.04193) |
| <a id="paper-2609.04355"></a>2609.04355<br>VLA | [VLA-Precision: Asymmetric Co-Bootstrapping for Efficient Real-World Online RL of Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.04355.pdf) | UR5e（单臂及双臂）；Franka FR3 | PGI-140-80夹爪；另有UR5e＋LinkerHand L20；[证据](research/robot_hardware/evidence.md#paper-2609.04355) |
| <a id="paper-2609.05266"></a>2609.05266<br>WAM | [TacPAC: Tactile Prediction and Real-Time Action Correction in World-Action Models for Contact-Rich Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.05266.pdf) | Flexiv（版本未明） | 双 InTac；[证据](research/robot_hardware/evidence.md#paper-2609.05266) |
| <a id="paper-2609.05282"></a>2609.05282<br>VLA | [Temporal Tactile Encoding and Compliance for Intent-Aware Robot-to-Human Bimanual Handover](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.05282.pdf) | 双臂平台（型号未明） | 人机交接；触觉与柔顺；[证据](research/robot_hardware/evidence.md#paper-2609.05282) |
| <a id="paper-2609.05376"></a>2609.05376<br>VLA | [What Matters, When? Diagnosing and Improving Conditional Visual Grounding in Visuomotor Imitation Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.05376.pdf) | UR3e | 条件视觉接地；[证据](research/robot_hardware/evidence.md#paper-2609.05376) |
| <a id="paper-2609.05588"></a>2609.05588<br>WAM | [GE-Act 2.0: Pretraining and Scaling a World-Action Model for Robotic Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.05588.pdf) | G1-OP；G2-90D（厂商待核） | 硬件配置名按原文保留；[证据](research/robot_hardware/evidence.md#paper-2609.05588) |
| <a id="paper-2609.05832"></a>2609.05832<br>VLA | [CR-VLA-Force: Learning Control-aware Compliance VLA Model for Robust Contact-rich Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.05832.pdf) | UR5e | 六维力传感；[证据](research/robot_hardware/evidence.md#paper-2609.05832) |
| <a id="paper-2609.05892"></a>2609.05892<br>VLA | [A4A: Cross-Embodiment Transfer of Action-Oriented 4D Affordances from Human Demonstrations](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.05892.pdf) | AgileX PiPER | 人视频 affordance 迁移；[证据](research/robot_hardware/evidence.md#paper-2609.05892) |
| <a id="paper-2609.06079"></a>2609.06079<br>VLA | [LayerRoute: Action-Conditioned Mixture-of-Layers Routing for Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.06079.pdf) | Franka（版本未明） | 三任务；[证据](research/robot_hardware/evidence.md#paper-2609.06079) |
| <a id="paper-2609.06114"></a>2609.06114<br>VLA | [Where Success Breaks: Failure-Boundary Learning for Robust Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.06114.pdf) | Franka FR3 | 失败边界；[证据](research/robot_hardware/evidence.md#paper-2609.06114) |
| <a id="paper-2609.06221"></a>2609.06221<br>VLA | [RefGuard: Identity-Aware Language-Guided Robot Manipulation via Joint Target-Anchor-Frame Grounding](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.06221.pdf) | UFACTORY UF850 | 身份感知抓取；[证据](research/robot_hardware/evidence.md#paper-2609.06221) |
| <a id="paper-2609.06251"></a>2609.06251<br>VLA | [MobileVLA-R1 2.0: RL-Enhanced Reasoning for Mobile Robot Control](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.06251.pdf) | Unitree Go2 | 移动导航；[证据](research/robot_hardware/evidence.md#paper-2609.06251) |
| <a id="paper-2609.06256"></a>2609.06256<br>VLA | [GloVLA: Let Geometry Move and Local VLA Interact for Robust Object-Centric Manipulation in Unstructured Environments](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.06256.pdf) | UR10e | 几何与局部 VLA；[证据](research/robot_hardware/evidence.md#paper-2609.06256) |
| <a id="paper-2609.06508"></a>2609.06508<br>VLA | [VLA-Corrector: Stage-Aware Observable State Understanding for Prompt-Based Closed-Loop Recovery of Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.06508.pdf) | 型号未明 | 目标位移恢复；[证据](research/robot_hardware/evidence.md#paper-2609.06508) |
| <a id="paper-2609.06578"></a>2609.06578<br>WAM | [Learning to Use Imagination: Progress-Conditioned Future Utilization for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.06578.pdf) | Galaxea R1 Lite；AgileX Cobot Magic | 两平台；[证据](research/robot_hardware/evidence.md#paper-2609.06578) |
| <a id="paper-2609.06852"></a>2609.06852<br>VLA | [ContextFlow: In-Context Flow Matching for Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.06852.pdf) | ALOHA（臂型号未明） | 单臂/双臂配置；[证据](research/robot_hardware/evidence.md#paper-2609.06852) |
| <a id="paper-2609.07006"></a>2609.07006<br>VLA | [GIFT: Goal-Injected Fine-Tuning for Efficient Manipulation Policy Adaptation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.07006.pdf) | Synria Alicia-D | D405观测；[证据](research/robot_hardware/evidence.md#paper-2609.07006) |
| <a id="paper-2609.07398"></a>2609.07398<br>WAM | [OpenWAM: An Open, Modular Exploration Towards Systematic World-Action Model Pretraining](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.07398.pdf) | Wuji 手＋Tianji 机械臂；RoboDojo 平台（待补） | 跨本体评测；[证据](research/robot_hardware/evidence.md#paper-2609.07398) |
| <a id="paper-2609.07581"></a>2609.07581<br>VLA | [ICI-VLA: In-Context Imitation with Spatiotemporally Aligned Demonstrations for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.07581.pdf) | ALOHA（臂型号未明） | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.07581) |
| <a id="paper-2609.08224"></a>2609.08224<br>VLA | [3DWay: Generalizing Robot Manipulation via 3D Consistent Waypoints](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.08224.pdf) | AgileX PiPER | KUKA 等跨本体项需区分仿真；[证据](research/robot_hardware/evidence.md#paper-2609.08224) |
| <a id="paper-2609.09119"></a>2609.09119<br>VLA | [DeCAL: Towards Physically-Grounded Dexterous Vision-Language-Action Models via Contact-Aware Latent Co-Imagination](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.09119.pdf) | UR5 ×2＋SharpaWave ×2 | 双手共56自由度；[证据](research/robot_hardware/evidence.md#paper-2609.09119) |
| <a id="paper-2609.09158"></a>2609.09158<br>VLA | [TANGO: Humanoid Navigation in Cluttered Environments with a Whole-Body Vision-Language-Action Model](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.09158.pdf) | Unitree G1 | 全身导航；[证据](research/robot_hardware/evidence.md#paper-2609.09158) |
| <a id="paper-2609.09808"></a>2609.09808<br>VLA | [GTA-2: A Multi-VLM Framework for Synthesizing Robot Manipulation Skills via Grounded Task Axes](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.09808.pdf) | UR5e；Franka（版本未明，基线） | 不同基线非同机对比；[证据](research/robot_hardware/evidence.md#paper-2609.09808) |
| <a id="paper-2609.09925"></a>2609.09925<br>VLA | [Time-Frequency Geometric Cross-Attention for Chunked Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.09925.pdf) | AgiBot A2 | 时频动作；[证据](research/robot_hardware/evidence.md#paper-2609.09925) |
| <a id="paper-2609.09941"></a>2609.09941<br>VLA | [HaWMPO: Hallucination-Aware World Model-based Policy Optimization for Generalist Robot Policy](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.09941.pdf) | G1（厂商待核） | 两任务；[证据](research/robot_hardware/evidence.md#paper-2609.09941) |
| <a id="paper-2609.10021"></a>2609.10021<br>VLA | [RoboDrop: Curating VLA Post-Training Data via Local Gradient Compatibility](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.10021.pdf) | ARX（双臂，版本未明） | 数据筛选；[证据](research/robot_hardware/evidence.md#paper-2609.10021) |
| <a id="paper-2609.10243"></a>2609.10243<br>WAM/VLA | [FolDeX: A Physical-World Benchmark for Long-Horizon Robotic Manipulation of Deformable Objects](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.10243.pdf) | 评测平台型号未明 | 含实体折衣试验；数据覆盖ALOHA、YAM、PiPER、Astribot、Franka、UR5、AgiBot G2及自制移动双臂，不能全部当作本次部署型号；[证据](research/robot_hardware/evidence.md#paper-2609.10243) |
| <a id="paper-2609.10405"></a>2609.10405<br>VLA | [Frequency-Conditioned Flow Matching for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.10405.pdf) | AgiBot Expedition A2 | 六任务；[证据](research/robot_hardware/evidence.md#paper-2609.10405) |
| <a id="paper-2609.10706"></a>2609.10706<br>VLA | [HuRo: Robotizing Human Videos for Scalable VLA Pretraining](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.10706.pdf) | ALLEX | 人形/灵巧平台；代际未明；[证据](research/robot_hardware/evidence.md#paper-2609.10706) |
| <a id="paper-2609.10915"></a>2609.10915<br>VLA | [IMLE-VLA: Fast Single-Step Action Generation for Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.10915.pdf) | Franka（版本未明） | 单步动作；[证据](research/robot_hardware/evidence.md#paper-2609.10915) |
| <a id="paper-2609.11561"></a>2609.11561<br>WAM | [Memory as Plans: World-Action Modeling with Memory-Grounded Planning](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.11561.pdf) | Franka FR3 | 记忆规划；[证据](research/robot_hardware/evidence.md#paper-2609.11561) |
| <a id="paper-2609.11697"></a>2609.11697<br>WAM/VLA | [ActSafeGuard: Differentiable and Training-Aligned Constraint Enforcement for Flow-Matching Policies](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.11697.pdf) | AgileX ALOHA（臂型号未明） | 双臂真实约束执行；[证据](research/robot_hardware/evidence.md#paper-2609.11697) |
| <a id="paper-2609.11875"></a>2609.11875<br>VLA | [UniMPA: A Unified Memory-Prediction-Action Model via Action-Grounded Transition Modeling](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.11875.pdf) | AgileX Cobot Magic；Galaxea R1 Lite | 两平台；[证据](research/robot_hardware/evidence.md#paper-2609.11875) |
| <a id="paper-2609.12316"></a>2609.12316<br>VLA | [DATAFARM: Distribution-Aligned Task and Motion Planning for Fine-Tuning Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.12316.pdf) | Franka Panda | DROID；Robotiq 2F-85；[证据](research/robot_hardware/evidence.md#paper-2609.12316) |
| <a id="paper-2609.13053"></a>2609.13053<br>VLA | [Dynin-Robotics: Omnimodal Unified Diffusion Vision-Language-Action Model](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.13053.pdf) | Franka FR3 | 四设定；[证据](research/robot_hardware/evidence.md#paper-2609.13053) |
| <a id="paper-2609.13679"></a>2609.13679<br>VLA | [How to Better Train VLAs: Lessons Learned From the REAL-I Challenge at ICRA 2026](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.13679.pdf) | Leju Kuavo（版本未明） | REAL-I 现场挑战；[证据](research/robot_hardware/evidence.md#paper-2609.13679) |
| <a id="paper-2609.13695"></a>2609.13695<br>VLA | [GROOVE: Geometry-Guided Reduction of Operational-Space Jerk in VLA Execution](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.13695.pdf) | UR5e | 配对平滑执行；[证据](research/robot_hardware/evidence.md#paper-2609.13695) |
| <a id="paper-2609.13812"></a>2609.13812<br>VLA | [GeomVLA: Unifying Scene, Motion, and Action in 3D](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.13812.pdf) | ALOHA 2 | 臂细节以本文为准；[证据](research/robot_hardware/evidence.md#paper-2609.13812) |
| <a id="paper-2609.13851"></a>2609.13851<br>VLA | [ReWeight: Leveraging Human Data for VLA Post-Training via Demonstration Retrieval and Sample Weighting](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.13851.pdf) | Dobot（双臂，版本未明） | 人数据后训练；[证据](research/robot_hardware/evidence.md#paper-2609.13851) |
| <a id="paper-2609.13984"></a>2609.13984<br>VLA | [What Makes an Efficient VLA? Navigating Action-Head Design, Scaling, and Latency](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.13984.pdf) | SO-ARM101 | 动作头对比；[证据](research/robot_hardware/evidence.md#paper-2609.13984) |
| <a id="paper-2609.14156"></a>2609.14156<br>VLA | [Visible Touch: Rendering Contact for Visuomotor Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.14156.pdf) | xArm7 | BOTA＋自制触觉；区分传感器标定与策略实验；[证据](research/robot_hardware/evidence.md#paper-2609.14156) |
| <a id="paper-2609.14219"></a>2609.14219<br>VLA | [Task-Specified Active Metrological Inspection with Measurement-Steered VLA Manipulation and Deterministic Evidence Gating](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.14219.pdf) | xArm6 ×2 | Keyence LJ-X8200 测量臂＋操作臂；[证据](research/robot_hardware/evidence.md#paper-2609.14219) |
| <a id="paper-2609.15012"></a>2609.15012<br>VLA | [Atomic Motion Coordinate for Language-Steerable and Force-Responsive Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.15012.pdf) | 双臂平台（型号未明） | 语言及力响应；[证据](research/robot_hardware/evidence.md#paper-2609.15012) |
| <a id="paper-2609.15382"></a>2609.15382<br>WAM | [From Prediction to Decision: World-Model-Guided Action Selection for Continuous Pile Excavation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.15382.pdf) | XCMG XC958EV | 5.5 t 电动轮式装载机；[证据](research/robot_hardware/evidence.md#paper-2609.15382) |
| <a id="paper-2609.15570"></a>2609.15570<br>WAM | [DIDO: Distilling Interaction-Centric Dynamics into One-Step Denoising for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.15570.pdf) | Galbot G1 | 不是 Unitree/AgiBot G1；[证据](research/robot_hardware/evidence.md#paper-2609.15570) |
| <a id="paper-2609.16437"></a>2609.16437<br>VLA | [XRoboToolKit-T: Teleoperation with High Stability and Precision with Tactile Sensing for Contact-rich Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.16437.pdf) | UR5e ×2 | 夹爪版与灵巧手版；遥操作；[证据](research/robot_hardware/evidence.md#paper-2609.16437) |
| <a id="paper-2609.16503"></a>2609.16503<br>VLA | [Dense to MoE Adaptation for Compact Vision Language Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.16503.pdf) | 型号未明 | 紧凑 MoE 真机评测；[证据](research/robot_hardware/evidence.md#paper-2609.16503) |
| <a id="paper-2609.16641"></a>2609.16641<br>VLA | [SAVLA: Symmetry-Aware Vision-Language-Action Models for Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.16641.pdf) | SO-101 | 对称性；[证据](research/robot_hardware/evidence.md#paper-2609.16641) |
| <a id="paper-2609.16644"></a>2609.16644<br>WAM | [WholeBodyWAM: Generalizing Pre-trained World-Action Priors to Humanoid Loco-Manipulation via WBC-Grounded Coordination](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.16644.pdf) | Unitree G1 | BrainCo Revo 2 双手；[证据](research/robot_hardware/evidence.md#paper-2609.16644) |
| <a id="paper-2609.16864"></a>2609.16864<br>VLA | [TEMPO: Learning Temporal Context for Dynamic Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.16864.pdf) | 型号未明 | 动态操作；[证据](research/robot_hardware/evidence.md#paper-2609.16864) |
| <a id="paper-2609.17021"></a>2609.17021<br>VLA | [sensVLA: Spatially-Grounded Vision-Language-Action Model for Autonomous Wheel Loader](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.17021.pdf) | 轮式装载机（型号待核） | 工程车辆；[证据](research/robot_hardware/evidence.md#paper-2609.17021) |
| <a id="paper-2609.17035"></a>2609.17035<br>VLA | [SWIM: Vision-Language-Grounded Soft Whole-Body Interactive Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.17035.pdf) | 自制软体机器人 | 开环下发，不是在线 VLA 闭环；[证据](research/robot_hardware/evidence.md#paper-2609.17035) |
| <a id="paper-2609.17099"></a>2609.17099<br>WAM | [GeoLAM: Learning Geometry-Grounded Latent Actions from Unlabeled Human Videos](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.17099.pdf) | AgileX PiPER | 几何潜在动作；[证据](research/robot_hardware/evidence.md#paper-2609.17099) |
| <a id="paper-2609.17115"></a>2609.17115<br>VLA | [Intrinsic Robot Rewarding: Reusing VLA Representations for Autonomous Evaluation and Policy Improvement](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.17115.pdf) | COMAU Racer 3 | Robotiq Hand-E；[证据](research/robot_hardware/evidence.md#paper-2609.17115) |
| <a id="paper-2609.17210"></a>2609.17210<br>WAM/VLA | [FluxVLA Engine: A One-Stop VLA Engineering Platform for Embodied Intelligence](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.17210.pdf) | ALOHA（臂型号未明）；Oli 移动平台 | 工程平台；[证据](research/robot_hardware/evidence.md#paper-2609.17210) |
| <a id="paper-2609.17372"></a>2609.17372<br>WAM | [XPACE: Joint World and Action Modeling from Heterogeneous Experience](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.17372.pdf) | 型号未明 | 三项真实任务；[证据](research/robot_hardware/evidence.md#paper-2609.17372) |
| <a id="paper-2609.17524"></a>2609.17524<br>WAM | [Modality-Autoregressive World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.17524.pdf) | 型号未明 | RAIDEN 是遥操作工具，不是机器人型号；[证据](research/robot_hardware/evidence.md#paper-2609.17524) |
| <a id="paper-2609.18016"></a>2609.18016<br>WAM | [Causal-History Test-Time Scaling for Failure Recovery in Autoregressive World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18016.pdf) | SO-101 ×2 | 历史恢复；[证据](research/robot_hardware/evidence.md#paper-2609.18016) |
| <a id="paper-2609.18084"></a>2609.18084<br>VLA | [Not All Layers Need Tuning: Diagnosing and Directing Adaptation in Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18084.pdf) | xArm（版本未明） | 参数适配；[证据](research/robot_hardware/evidence.md#paper-2609.18084) |
| <a id="paper-2609.18108"></a>2609.18108<br>VLA | [Technical Report: One-Step Drifting Action Heads for GR00T N1.7](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18108.pdf) | LimX TRON2 | 初步实体部署与采集；没有受控成功率对比，不等同受控真机基准；[证据](research/robot_hardware/evidence.md#paper-2609.18108) |
| <a id="paper-2609.18197"></a>2609.18197<br>WAM | [WholeBodyWAM: Learning Whole-Body World Action Models with Scalable Motion Priors](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18197.pdf) | TianGong 3.0 | 全尺寸人形；[证据](research/robot_hardware/evidence.md#paper-2609.18197) |
| <a id="paper-2609.18207"></a>2609.18207<br>VLA | [Reinforcement Learning for Real-Time Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18207.pdf) | 型号未明 | 实时强化学习；[证据](research/robot_hardware/evidence.md#paper-2609.18207) |
| <a id="paper-2609.18242"></a>2609.18242<br>VLA | [ForceDelta-VLA: Distilling Force-Conditioned ActionCorrections for Contact-Rich Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18242.pdf) | Franka Panda；X Square Robot（版本未明） | 单臂及双臂；[证据](research/robot_hardware/evidence.md#paper-2609.18242) |
| <a id="paper-2609.18243"></a>2609.18243<br>WAM/VLA | [Acting in Meters: Learning Metric Interactions for Precise Robotic Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18243.pdf) | AgileX PiPER-H ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.18243) |
| <a id="paper-2609.18259"></a>2609.18259<br>VLA | [M2Tok: Multi-head Multi-codebook Discrete Action Tokenization for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18259.pdf) | AgileX Cobot Magic（PiPER） | 四臂含主臂和从臂；[证据](research/robot_hardware/evidence.md#paper-2609.18259) |
| <a id="paper-2609.18374"></a>2609.18374<br>WAM/VLA | [Decoupling Vision, Language, and Action for Efficient Multi-Task Robot Policies](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18374.pdf) | xArm7＋导轨 | 8-DoF；[证据](research/robot_hardware/evidence.md#paper-2609.18374) |
| <a id="paper-2609.18392"></a>2609.18392<br>WAM/VLA | [DistAL: Distance-based Advantage Learning for VLA Fine-Tuning](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18392.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.18392) |
| <a id="paper-2609.18462"></a>2609.18462<br>WAM | [CSWAM: Better Causal Semantic Representations for Out-of-Distribution Generalization in World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18462.pdf) | ARX（双臂，版本未明） | 分拣/折叠；[证据](research/robot_hardware/evidence.md#paper-2609.18462) |
| <a id="paper-2609.18497"></a>2609.18497<br>WAM/VLA | [TAO-Force: Unifying Force-Aware Perception and Fast-Slow Control for Contact-Rich Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18497.pdf) | AgiBot G1 | 双腕六维力；[证据](research/robot_hardware/evidence.md#paper-2609.18497) |
| <a id="paper-2609.18514"></a>2609.18514<br>VLA | [ActiveScale: Scaling Active Perception for Robots across Model, Data, and Hardware](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18514.pdf) | AMP 自制主动感知移动平台 | 机械臂型号待补；[证据](research/robot_hardware/evidence.md#paper-2609.18514) |
| <a id="paper-2609.18663"></a>2609.18663<br>VLA | [VLA-ULAP: Interleaving Cloud VLA Calls with Ultra-Lightweight Local Action Prediction at the Edge](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.18663.pdf) | SO-101 | 端云协同；[证据](research/robot_hardware/evidence.md#paper-2609.18663) |
| <a id="paper-2609.19104"></a>2609.19104<br>VLA | [rMuscle: Robotic Muscle Memory for Efficient Vision-Language-Action Model Inference](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.19104.pdf) | ALOHA；Dobot；Franka（具体版本未明） | 双臂摆放与双机器人流水线；[证据](research/robot_hardware/evidence.md#paper-2609.19104) |
| <a id="paper-2609.19441"></a>2609.19441<br>WAM | [Predict Before You Deploy: Offline Prediction of Quantization-Induced Task Degradation for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.19441.pdf) | Franka FR3 | 量化风险真机验证；[证据](research/robot_hardware/evidence.md#paper-2609.19441) |
| <a id="paper-2609.19579"></a>2609.19579<br>VLA | [Recovering Aggressively Pruned Vision-Language-Action Models with Offline Hidden-State Distillation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.19579.pdf) | AgileX PiPER | 剪枝后恢复；[证据](research/robot_hardware/evidence.md#paper-2609.19579) |
| <a id="paper-2609.19600"></a>2609.19600<br>VLA | [WorldContact: A Contact-Centric World Model for Scalable Robot Learning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.19600.pdf) | AgileX PiPER | 接触世界模型数据；[证据](research/robot_hardware/evidence.md#paper-2609.19600) |
| <a id="paper-2609.19613"></a>2609.19613<br>WAM | [TacSushi: Tactile-Grounded World-Action Modeling for Dexterous Sushi Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.19613.pdf) | Shadow Dexterous Hand Lite | 臂型号未明；STF 触觉；[证据](research/robot_hardware/evidence.md#paper-2609.19613) |
| <a id="paper-2609.19666"></a>2609.19666<br>VLA | [Towards High-DoF Dexterous Manipulation through VLA Post-Training](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.19666.pdf) | Tianji 7-DoF 臂 ×2＋Wuji 20-DoF 手 ×2 | 灵巧后训练；[证据](research/robot_hardware/evidence.md#paper-2609.19666) |
| <a id="paper-2609.19923"></a>2609.19923<br>VLA | [Co-VLA: Consensus-based Federated Training for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.19923.pdf) | SO-101 | 联邦策略闭环；[证据](research/robot_hardware/evidence.md#paper-2609.19923) |
| <a id="paper-2609.19974"></a>2609.19974<br>WAM | [MaskHarness-WAM: Instance-Grounded Harnessing for Long-Horizon Robot Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.19974.pdf) | ARX（双臂，版本未明） | 目标掩码长程任务；[证据](research/robot_hardware/evidence.md#paper-2609.19974) |
| <a id="paper-2609.19976"></a>2609.19976<br>VLA | [Compliance for Free: Learning Identifiable Impedance via Bilateral Teleoperation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.19976.pdf) | Franka FR3 | 双边遥操作柔顺；[证据](research/robot_hardware/evidence.md#paper-2609.19976) |
| <a id="paper-2609.20646"></a>2609.20646<br>VLA | [TraceFlow: Guiding Frozen Flow-Matching Robot Policies with Success and Failure Traces](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.20646.pdf) | ARX AC-One | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.20646) |
| <a id="paper-2609.20648"></a>2609.20648<br>VLA | [SkipVLA: Skipping VLA Steps with Classical Planning for Fast Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.20648.pdf) | I2RT YAM | 6DoF真实操作；[证据](research/robot_hardware/evidence.md#paper-2609.20648) |
| <a id="paper-2609.20659"></a>2609.20659<br>VLA | [HIL-UMI: Bringing Human-in-the-Loop Post-Training of Vision-Language-Action Models to Universal Manipulation Interface](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.20659.pdf) | Franka（版本未明） | UMI 接管后训练；[证据](research/robot_hardware/evidence.md#paper-2609.20659) |
| <a id="paper-2609.20709"></a>2609.20709<br>WAM | [MoWAM: Explicit Future Motion Prediction for Efficient World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.20709.pdf) | Franka FR3 | 未来运动预测；[证据](research/robot_hardware/evidence.md#paper-2609.20709) |
| <a id="paper-2609.20761"></a>2609.20761<br>WAM | [Agile-WAM: An Agile Tactile World Action Model for Contact-Rich Robot Control](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.20761.pdf) | Flexiv Rizon 4 | 五项接触任务；[证据](research/robot_hardware/evidence.md#paper-2609.20761) |
| <a id="paper-2609.20776"></a>2609.20776<br>VLA | [GeoAAC: Geometry-Based Adaptive Action Chunking from Denoising Trajectories in VLA Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.20776.pdf) | AgileX PiPER-X ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.20776) |
| <a id="paper-2609.20980"></a>2609.20980<br>VLA | [ForeTac-VLA: A Forecasting-Based Tactile-Vision-Language-Action Model for Contact-Rich Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.20980.pdf) | UR7e | Robotiq 2F-85；[证据](research/robot_hardware/evidence.md#paper-2609.20980) |
| <a id="paper-2609.21022"></a>2609.21022<br>VLA | [Catch Me If You Can: Real-Time Feedback Denoising for Responsive VLAs](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.21022.pdf) | Franka Panda | 静态/动态任务；[证据](research/robot_hardware/evidence.md#paper-2609.21022) |
| <a id="paper-2609.21228"></a>2609.21228<br>VLA | [FOCAL-VLA: Subtask-Guided Geometry Distillation and Implicit World Modeling for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.21228.pdf) | Franka Panda | 固定外置相机；[证据](research/robot_hardware/evidence.md#paper-2609.21228) |
| <a id="paper-2609.21358"></a>2609.21358<br>VLA | [FAN: Foresight Action Normalization for Continual Adaptation of Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.21358.pdf) | AgileX PiPER-X ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.21358) |
| <a id="paper-2609.21449"></a>2609.21449<br>WAM | [ME-Dex 1.0: Bringing Heterogeneous Tactile Sensing into World Action Modeling](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.21449.pdf) | SO-101；Flex2 | 定性真实 rollout；不据名称映射 Flexiv；[证据](research/robot_hardware/evidence.md#paper-2609.21449) |
| <a id="paper-2609.21474"></a>2609.21474<br>WAM | [MT-WAM: Reorienting the One-Pass Predictive Representation Toward Action Generation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.21474.pdf) | UR5（单臂及双臂） | 四任务；[证据](research/robot_hardware/evidence.md#paper-2609.21474) |
| <a id="paper-2609.21514"></a>2609.21514<br>WAM | [Skel-WAM: A Hand-Skeleton-Conditioned World Action Model for Human-to-Robot Manipulation Transfer](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.21514.pdf) | Franka FR3 ×2＋RobotEra XHand1 ×2 | 骨架迁移；[证据](research/robot_hardware/evidence.md#paper-2609.21514) |
| <a id="paper-2609.21650"></a>2609.21650<br>VLA | [SynthDemo-RL: Breaking the Zero-Reward Barrier in VLA Adaptation with LLM-Guided Synthetic Demonstrations](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.21650.pdf) | WidowX（版本未明） | 真实开环动作回放，非闭环 sim2real；[证据](research/robot_hardware/evidence.md#paper-2609.21650) |
| <a id="paper-2609.21753"></a>2609.21753<br>VLA | [PSR: Predictive Sensorimotor Representation Learning for Contact-Rich Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.21753.pdf) | 型号未明 | 接触预测表征；[证据](research/robot_hardware/evidence.md#paper-2609.21753) |
| <a id="paper-2609.21817"></a>2609.21817<br>VLA | [A Sim-to-Real Integration Pipeline for Training and Deployment of Chunk-Based VLA Manipulation Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.21817.pdf) | Franka FR3 | 采集开环＋策略闭环；[证据](research/robot_hardware/evidence.md#paper-2609.21817) |
| <a id="paper-2609.21908"></a>2609.21908<br>VLA | [CommitFlow: Semantic Commitment Verification and Local Correction for Long-Horizon Robot Manipulation VLA Execution](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.21908.pdf) | 型号未明 | 两项真实任务；[证据](research/robot_hardware/evidence.md#paper-2609.21908) |
| <a id="paper-2609.21948"></a>2609.21948<br>VLA | [GALA: Geometry-Aware Latent Action Modeling for Vision-Language-Action Model Pretraining across Embodiments](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.21948.pdf) | 机械臂型号未明＋XHand 12DoF | 灵巧手名称可确认；[证据](research/robot_hardware/evidence.md#paper-2609.21948) |
| <a id="paper-2609.21983"></a>2609.21983<br>WAM | [SkelWAM: A Skeleton-Guided World-Action Model for Zero-Shot Cross-Embodiment Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.21983.pdf) | JAKA mini2；Feagine A03 | 真机跨本体；[证据](research/robot_hardware/evidence.md#paper-2609.21983) |
| <a id="paper-2609.22332"></a>2609.22332<br>WAM | [AffordanceWAM: Affordance-Aware Joint World-Action Modeling for Robot Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.22332.pdf) | 型号未明 | 真实操作；[证据](research/robot_hardware/evidence.md#paper-2609.22332) |
| <a id="paper-2609.22462"></a>2609.22462<br>VLA | [VLPSA: Vision-Language-Poisson-Safe Actions for Full-Body Safety of Learned Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.22462.pdf) | Franka FR3 | 全臂避障；[证据](research/robot_hardware/evidence.md#paper-2609.22462) |
| <a id="paper-2609.22606"></a>2609.22606<br>VLA | [VT-Bridge: Bridging Pretrained Foundation VLAs to VTLAs via Lightweight Residual Adaptation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.22606.pdf) | Franka Panda | 触觉残差；[证据](research/robot_hardware/evidence.md#paper-2609.22606) |
| <a id="paper-2609.22684"></a>2609.22684<br>VLA | [StateMem: Single-State Residual Memory with Adaptive Inference for Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.22684.pdf) | xArm7＋Inspire 灵巧手 | 记忆；[证据](research/robot_hardware/evidence.md#paper-2609.22684) |
| <a id="paper-2609.22840"></a>2609.22840<br>VLA | [ForceRFT: Refining VLA Actions through Force-Guided Residual Reinforcement Learning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.22840.pdf) | Franka FR3 | OnRobot HEX-E；Robotiq 2F-85；[证据](research/robot_hardware/evidence.md#paper-2609.22840) |
| <a id="paper-2609.22879"></a>2609.22879<br>VLA | [Prioritized Rollouts for Efficient World Model-based Vision-Language-Action Policy Optimization](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.22879.pdf) | ARX-X5 ×2 | 优先级真实rollout；[证据](research/robot_hardware/evidence.md#paper-2609.22879) |
| <a id="paper-2609.22888"></a>2609.22888<br>VLA | [Stable and Efficient Real-World Online VLA Post-Training via Asynchronous Replay-Anchored Policy Improvement](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.22888.pdf) | Franka FR3 | 真实在线后训练；[证据](research/robot_hardware/evidence.md#paper-2609.22888) |
| <a id="paper-2609.22895"></a>2609.22895<br>VLA | [H-VLA: Hierarchical Vision-Language-Action Model with Key-Action Reasoning and Motion Planning in a Unified Action Space](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.22895.pdf) | AgileX Cobot Magic-style（臂型号未明） | 保留论文平台表述；[证据](research/robot_hardware/evidence.md#paper-2609.22895) |
| <a id="paper-2609.22973"></a>2609.22973<br>VLA | [An Empirical Study and Open Testbed for Federated Fine-Tuning of Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.22973.pdf) | SO-101 | 联邦测试床；[证据](research/robot_hardware/evidence.md#paper-2609.22973) |
| <a id="paper-2609.23133"></a>2609.23133<br>VLA | [AquaCap: A Training-Free Underwater Embodied Agent with Code-as-Policy](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.23133.pdf) | 水下操作机器人（型号未明） | 水池/水下真机；[证据](research/robot_hardware/evidence.md#paper-2609.23133) |
| <a id="paper-2609.23275"></a>2609.23275<br>VLA | [SCULPT-VLA: Learning Structured Control through Staged Action Grounding](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.23275.pdf) | UR5e | 三指柔性夹爪；[证据](research/robot_hardware/evidence.md#paper-2609.23275) |
| <a id="paper-2609.23565"></a>2609.23565<br>VLA | [MaskVLA: Visual Masking Against Trajectory Overfitting of Vision-Language-Action Model](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.23565.pdf) | AIRBOT Mobile ALOHA | 主从各6-DoF＋夹爪；[证据](research/robot_hardware/evidence.md#paper-2609.23565) |
| <a id="paper-2609.23578"></a>2609.23578<br>WAM/VLA | [AR-WAM: A Visual-Conditioned Agent-Ready World Action Model for Robotic Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.23578.pdf) | Astribot S1 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.23578) |
| <a id="paper-2609.23614"></a>2609.23614<br>VLA | [CompVLA: A Variable Compliance Vision-Language-Action Model for Contact-rich Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.23614.pdf) | Franka FR3 ×2 | 双边柔顺；[证据](research/robot_hardware/evidence.md#paper-2609.23614) |
| <a id="paper-2609.23650"></a>2609.23650<br>VLA | [Beyond Appearance Shifts: Task-Semantic Action Calibration for VLA Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.23650.pdf) | AgileX PiPER | 任务语义校准；[证据](research/robot_hardware/evidence.md#paper-2609.23650) |
| <a id="paper-2609.23863"></a>2609.23863<br>WAM/VLA | [Grounded Action Model: 3D Grounding as a Foundation for Robotics](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.23863.pdf) | YAM ×2；Franka（版本未明） | 点/框接口；[证据](research/robot_hardware/evidence.md#paper-2609.23863) |
| <a id="paper-2609.23888"></a>2609.23888<br>WAM | [HapticWAM: Distilling Imagined Touch into a World-Action Model without Inference-Time Tactile Sensing](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.23888.pdf) | UR3 | Robotiq 2F-85；DM-Tac W2L；[证据](research/robot_hardware/evidence.md#paper-2609.23888) |
| <a id="paper-2609.23910"></a>2609.23910<br>VLA | [ReVeal: A Reconstruction-Aware Real-to-Sim Framework for VLA Policy Evaluation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.23910.pdf) | 型号未明 | 真实对仿真相关性；[证据](research/robot_hardware/evidence.md#paper-2609.23910) |
| <a id="paper-2609.23944"></a>2609.23944<br>VLA | [Topology-Informed Visual Prompting For Vision Language Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.23944.pdf) | 型号未明 | 真实提箱；[证据](research/robot_hardware/evidence.md#paper-2609.23944) |
| <a id="paper-2609.23968"></a>2609.23968<br>VLA | [Opt2VLA: Force-Aware Vision-Language-Action for Contact-Rich Humanoid Whole-Body Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.23968.pdf) | 人形平台（型号未明） | 力响应全身控制；[证据](research/robot_hardware/evidence.md#paper-2609.23968) |
| <a id="paper-2609.24033"></a>2609.24033<br>VLA | [Imagine-RL: Residual-Confidence-Guided Cross-Attention for World-Model-Augmented VLA Reinforcement Learning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.24033.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.24033) |
| <a id="paper-2609.24118"></a>2609.24118<br>VLA | [CARE: Experience-Guided Atomic Corrective Execution for Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.24118.pdf) | SO-101 ×2 | 双臂纠正执行；[证据](research/robot_hardware/evidence.md#paper-2609.24118) |
| <a id="paper-2609.24187"></a>2609.24187<br>VLA | [StenoVLA-3D: 3D-Aware Reasoning VLA for Navigation Through Gastrointestinal Stenoses](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.24187.pdf) | 机器人内镜（型号未明） | 狭窄消化道实体平台待详细核对；[证据](research/robot_hardware/evidence.md#paper-2609.24187) |
| <a id="paper-2609.24350"></a>2609.24350<br>WAM/VLA | [LIBERO-VPro: Benchmarking Closed-Loop Visual Robustness of Robotic Foundation Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.24350.pdf) | Franka FR3 | 视觉鲁棒性；[证据](research/robot_hardware/evidence.md#paper-2609.24350) |
| <a id="paper-2609.24433"></a>2609.24433<br>VLA | [FoldQuantVLA: Native Low-Bit Quantization of Vision-Language-Action Models via Consistent Folding](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.24433.pdf) | ALOHA（单臂）；SO-101 | 量化部署；[证据](research/robot_hardware/evidence.md#paper-2609.24433) |
| <a id="paper-2609.24525"></a>2609.24525<br>VLA | [Bridge3D: Enabling Vision-Language-Action Models to See and Act in 3D](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/extracts/2609.24525.txt) | KEENON XMAN-R1；AgileX PiPER | 主平台＋空间泛化补充；[证据](research/robot_hardware/evidence.md#paper-2609.24525) |
| <a id="paper-2609.24682"></a>2609.24682<br>VLA | [Think Like a World Model, Act Like a VLA: Distilling World-Model Representations into Compact Robot Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.24682.pdf) | AgileX Nero；TRIP-Bag | 单臂与7-DoF双臂；[证据](research/robot_hardware/evidence.md#paper-2609.24682) |
| <a id="paper-2609.24868"></a>2609.24868<br>WAM | [DualWAM: Dual-System World Action Models for Asynchronous Global Planning and Local Refinement](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.24868.pdf) | Galbot G1；Franka FR3 | 各十项真实任务；[证据](research/robot_hardware/evidence.md#paper-2609.24868) |
| <a id="paper-2609.24976"></a>2609.24976<br>WAM | [DexTacWAM: A Visuo-Tactile World-Action Model for Dexterous Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.24976.pdf) | 双臂＋Sharpa 22-DoF 手（臂型号未明） | 灵巧触觉；[证据](research/robot_hardware/evidence.md#paper-2609.24976) |
| <a id="paper-2609.25187"></a>2609.25187<br>VLA | [X-Planner: Event-Structured Task Planning for Embodied Intelligence](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.25187.pdf) | 型号未明 | 事件规划；[证据](research/robot_hardware/evidence.md#paper-2609.25187) |
| <a id="paper-2609.25369"></a>2609.25369<br>VLA | [Capability-Aware Arbitration for Semantic Intent-Based Shared Control](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.25369.pdf) | SO-101 | Joy-Con 共享控制；[证据](research/robot_hardware/evidence.md#paper-2609.25369) |
| <a id="paper-2609.25376"></a>2609.25376<br>VLA | [VLAQuantBench: Closed-Loop Evaluation of Post-Training Quantization for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.25376.pdf) | Franka FR3 | Franka Hand；量化评测；[证据](research/robot_hardware/evidence.md#paper-2609.25376) |
| <a id="paper-2609.25606"></a>2609.25606<br>VLA | [CableVLA: Simulation-Privileged Global-Local Representation Learning for Cable Routing](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.25606.pdf) | RealMan RM75-6F | Inspire EG2-4C2；Loong96触觉；[证据](research/robot_hardware/evidence.md#paper-2609.25606) |
| <a id="paper-2609.25627"></a>2609.25627<br>WAM/VLA | [MachEmbodied-U0: Unified Understanding and Generation Model for Embodied Intelligence](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.25627.pdf) | 型号未明 | 定性真实部署；[证据](research/robot_hardware/evidence.md#paper-2609.25627) |
| <a id="paper-2609.25750"></a>2609.25750<br>VLA | [Fisheye-VLA: Decoupling Coverage and Acuity for Manipulation with a Single Fisheye Camera](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.25750.pdf) | Flexiv Rizon 4 ×2 | GN01；鱼眼相机；[证据](research/robot_hardware/evidence.md#paper-2609.25750) |
| <a id="paper-2609.25756"></a>2609.25756<br>VLA | [MedVLA: A Hierarchical Vision-Language-Action Framework for Closed-Loop Precision Medical Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.25756.pdf) | 精密医疗机械平台（型号未明） | 实体插入；不等同临床手术；[证据](research/robot_hardware/evidence.md#paper-2609.25756) |
| <a id="paper-2609.25785"></a>2609.25785<br>VLA | [VisForce: Visual Grounding of Current and Desired Forces for Goal-Conditioned Dexterous Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.25785.pdf) | UR10＋Inspire RH56F1 | 力接地；UR10 不补 e；[证据](research/robot_hardware/evidence.md#paper-2609.25785) |
| <a id="paper-2609.26313"></a>2609.26313<br>VLA | [SafeLoop: Risk-Aware Rollback for Vision-Language-Action Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.26313.pdf) | AgileX PiPER ×2 | 双臂风险回滚；[证据](research/robot_hardware/evidence.md#paper-2609.26313) |
| <a id="paper-2609.26467"></a>2609.26467<br>VLA | [RouteRLT: Learning When and Which RL Specialist Should Control a Vision-Language-Action Policy](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.26467.pdf) | Trossen / ALOHA（臂型号未明） | 插入前有人工对齐；[证据](research/robot_hardware/evidence.md#paper-2609.26467) |
| <a id="paper-2609.26672"></a>2609.26672<br>VLA | [Imperfection for Precision: Upcycling Imperfect Data for High-Precision Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.26672.pdf) | Rainbow Robotics RB-Y1 | 精密操作；[证据](research/robot_hardware/evidence.md#paper-2609.26672) |
| <a id="paper-2609.26868"></a>2609.26868<br>VLA | [Backdoors in Learning-Based Industrial Robotic Arm Manipulation: An Empirical Security Study](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.26868.pdf) | FANUC LR Mate 200iD；xArm6 | 物理攻击实验；[证据](research/robot_hardware/evidence.md#paper-2609.26868) |
| <a id="paper-2609.27314"></a>2609.27314<br>WAM | [CoRe-WAM: Correspondence-Aligned Temporal Residuals for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.27314.pdf) | Astribot S1 | 时间残差；[证据](research/robot_hardware/evidence.md#paper-2609.27314) |
| <a id="paper-2609.27450"></a>2609.27450<br>VLA | [BEE: Intervention-Adaptive Real-World Reinforcement Learning with Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.27450.pdf) | 型号未明 | 真实干预自适应RL；[证据](research/robot_hardware/evidence.md#paper-2609.27450) |
| <a id="paper-2609.27455"></a>2609.27455<br>WAM | [Latent evolving World Action Model](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.27455.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.27455) |
| <a id="paper-2609.27468"></a>2609.27468<br>VLA | [CereVLA: Cerebellum-Inspired Consequence-Aware Residual Governance for Efficient Vision-Language-Action Execution](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.27468.pdf) | SO-101 | 残差控制；[证据](research/robot_hardware/evidence.md#paper-2609.27468) |
| <a id="paper-2609.27734"></a>2609.27734<br>VLA | [InfiNoVA: Infinite Novel View Augmentation for Viewpoint Invariant Robot Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.27734.pdf) | SO-101 | 视角增强；[证据](research/robot_hardware/evidence.md#paper-2609.27734) |
| <a id="paper-2609.28161"></a>2609.28161<br>VLA | [Dissecting Advantage-Guided Post-Training for Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.28161.pdf) | 型号未明 | 优势后训练；[证据](research/robot_hardware/evidence.md#paper-2609.28161) |
| <a id="paper-2609.28256"></a>2609.28256<br>VLA | [MemBodied: Recurrent Associative Memory for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.28256.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.28256) |
| <a id="paper-2609.28314"></a>2609.28314<br>VLA | [TANDEM: Task and Motion Planning with As-Needed Demonstrations for Efficient Vision-Language-Action Model Fine-tuning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.28314.pdf) | 机械臂型号未明 | 真实TAMP执行、人工干预及π0.5-DROID微调；不能由DROID检查点补Panda型号；[证据](research/robot_hardware/evidence.md#paper-2609.28314) |
| <a id="paper-2609.28431"></a>2609.28431<br>WAM/VLA | [LiMA: Bridging Long-term Imagination to Real-time Dexterous Manipulation via Asynchronous Diffusion](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.28431.pdf) | UR5 ×2＋SharpaWave ×2 | 灵巧异步推理；[证据](research/robot_hardware/evidence.md#paper-2609.28431) |
| <a id="paper-2609.28811"></a>2609.28811<br>WAM | [DeltaWAM: Delta World Action Models for Bimanual Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.28811.pdf) | RealMan RM75-6F ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.28811) |
| <a id="paper-2609.28865"></a>2609.28865<br>VLA | [Direction-Scale Decomposition in Action Representation: Rethinking What to Tokenize for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.28865.pdf) | Franka（版本未明） | 方向尺度分解；[证据](research/robot_hardware/evidence.md#paper-2609.28865) |
| <a id="paper-2609.28927"></a>2609.28927<br>WAM | [Streaming-WAM: Action-Conditioned World-Action Model for Asynchronous Robot Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.28927.pdf) | 型号未明 | 盖章及积木；异步执行；[证据](research/robot_hardware/evidence.md#paper-2609.28927) |
| <a id="paper-2609.28955"></a>2609.28955<br>VLA | [ActGaze: Learning Action-Grounded Gaze through Counterfactual Visual Interventions for High-Precision Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.28955.pdf) | 机器人型号未明 | 四项高精度任务；[证据](research/robot_hardware/evidence.md#paper-2609.28955) |
| <a id="paper-2609.29204"></a>2609.29204<br>VLA | [AdaHVLA: Adaptive Harnesses for Long-Horizon Vision-Language-Action Execution](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.29204.pdf) | Unitree Go2 | 长程导航；[证据](research/robot_hardware/evidence.md#paper-2609.29204) |
| <a id="paper-2609.30092"></a>2609.30092<br>VLA | [Self-Adaptive VLA for Robust Robot Deployment](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.30092.pdf) | AgileX PiPER；Wuji 20-DoF 手 | 硬件变化自适应；手的搭载臂待核；[证据](research/robot_hardware/evidence.md#paper-2609.30092) |
| <a id="paper-2609.30247"></a>2609.30247<br>WAM | [Rolling-WAM: World Action Models with Rolling Imagination](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.30247.pdf) | Unitree G1 | 三项操作；[证据](research/robot_hardware/evidence.md#paper-2609.30247) |
| <a id="paper-2609.30833"></a>2609.30833<br>VLA | [Fast Plans, Faithful Actions: Closing the Planning-Execution Gap in Hierarchical Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.30833.pdf) | 型号未明 | 层次规划执行；[证据](research/robot_hardware/evidence.md#paper-2609.30833) |
| <a id="paper-2609.30868"></a>2609.30868<br>VLA | [VLaRL: Augmenting Vision-Language-Action Models with Simulation-Trained Latent-Conditioned Residual RL](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.30868.pdf) | Franka FR3 | 仿真残差直接迁移；[证据](research/robot_hardware/evidence.md#paper-2609.30868) |
| <a id="paper-2609.30913"></a>2609.30913<br>VLA | [Causeway: Restoring Task Accessibility for Instruction Switching in VLA Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.30913.pdf) | UFACTORY xArm6 | 指令切换真实评估；[证据](research/robot_hardware/evidence.md#paper-2609.30913) |
| <a id="paper-2609.30965"></a>2609.30965<br>VLA | [FRAM: Trajectory-Guided Visual Feature Selection for Compact Language-Conditioned Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.30965.pdf) | UR5e ×2 | 双臂叠杯；[证据](research/robot_hardware/evidence.md#paper-2609.30965) |
| <a id="paper-2609.31394"></a>2609.31394<br>WAM | [InternW0-$Δ$: A World Action Model Bridging Predictive Dynamics and Actions with 20K+ Hours of Open Data](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.31394.pdf) | ARX AC-One；ARX-5；Franka＋XHand；Tianji Marvin＋Wuji 手 | 四种真实配置；[证据](research/robot_hardware/evidence.md#paper-2609.31394) |
| <a id="paper-2609.31717"></a>2609.31717<br>VLA | [PanoFuse: Panorama-Enhanced Vision-Language-Action Learning with Decoupled Semantic-Geometric Routing](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.31717.pdf) | AgileX PiPER | U-ARM 作为主控接口；[证据](research/robot_hardware/evidence.md#paper-2609.31717) |
| <a id="paper-2609.31904"></a>2609.31904<br>VLA | [GT-VLA: Target-Conditioned Trace Guidance for Generalizable Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.31904.pdf) | I2RT YAM | GT-VLA；桌面长程任务；[证据](research/robot_hardware/evidence.md#paper-2609.31904) |
| <a id="paper-2609.32069"></a>2609.32069<br>VLA | [Find Something You Can’t Do: Agentic Real-World Reinforcement Learning for Self-Improving VLA Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.32069.pdf) | Franka Panda | 自主课程RL；[证据](research/robot_hardware/evidence.md#paper-2609.32069) |
| <a id="paper-2609.32108"></a>2609.32108<br>VLA | [SAMBAR: Selective Anchoring via Method of Multipliers for Balanced Knowledge Acquisition and Retention in Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.32108.pdf) | xArm7 | 新旧任务适配；[证据](research/robot_hardware/evidence.md#paper-2609.32108) |
| <a id="paper-2609.32155"></a>2609.32155<br>VLA | [RECASTVLA: FROM PAST INTERACTION TO FUTURE CONTROL WITH ADAPTIVE POLICY STATES](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.32155.pdf) | SO-ARM101 ×2 | 12项任务；每策略240次试验；[证据](research/robot_hardware/evidence.md#paper-2609.32155) |
| <a id="paper-2609.32193"></a>2609.32193<br>WAM/VLA | [Devol-ONE: One Autoregressive Mixture of Transformers to Unify Vision-Language-Action and Latent World Modeling](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.32193.pdf) | Flexiv Rizon 4；Flexiv Rizon 4R | 左右臂；单臂和双臂设置；[证据](research/robot_hardware/evidence.md#paper-2609.32193) |
| <a id="paper-2609.32634"></a>2609.32634<br>VLA | [PF-RL: PROGRESS FIELD REINFORCEMENT LEARNING VIA GOAL-CONDITIONED VALUE GEOMETRY FOR VISION-LANGUAGE-ACTION MODELS](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.32634.pdf) | AgileX Cobot Magic（PiPER ×2） | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.32634) |
| <a id="paper-2609.32767"></a>2609.32767<br>VLA | [CLAP: Closed-Loop Alignment with Pressure for Precise Suction Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.32767.pdf) | Unitree Z1 | 吸盘压力反馈；[证据](research/robot_hardware/evidence.md#paper-2609.32767) |
| <a id="paper-2609.32779"></a>2609.32779<br>WAM | [Copper-Policy: Focus on the Representation for Robust Robot Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.32779.pdf) | AgileX Cobot（PiPER ×2） | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.32779) |
| <a id="paper-2609.33125"></a>2609.33125<br>VLA | [TRAIN TOGETHER OR MERGE LATER? UNIFYING VLA EXPERTS VIA A SHARED ACTION INTERFACE](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33125.pdf) | AgileX PiPER | 专家合并；[证据](research/robot_hardware/evidence.md#paper-2609.33125) |
| <a id="paper-2609.33157"></a>2609.33157<br>VLA | [TimelyDAgger: Timing-Aware Expert Querying for VLA Policy Improvement](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33157.pdf) | 双臂平台（型号未明） | V.C真实线缆交接及连接器插入；[证据](research/robot_hardware/evidence.md#paper-2609.33157) |
| <a id="paper-2609.33172"></a>2609.33172<br>WAM | [Dynamic Manipulation with World-Action Models via Counterfactual Planning](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33172.pdf) | I2RT YAM ×2 | 附录D.1主从双臂；动态目标；[证据](research/robot_hardware/evidence.md#paper-2609.33172) |
| <a id="paper-2609.33177"></a>2609.33177<br>WAM | [DeltaWAM: Change-Centric Visual Foresight via Delta Tokens for an Efficient World-Action Model](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33177.pdf) | AgileX PiPER | 定性执行；[证据](research/robot_hardware/evidence.md#paper-2609.33177) |
| <a id="paper-2609.33197"></a>2609.33197<br>VLA | [TAO-DA: Towards Autonomous Operation—A Dual-Arm Vision–Language–Action Model for Coordinated Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33197.pdf) | AgiBot G1 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.33197) |
| <a id="paper-2609.33256"></a>2609.33256<br>VLA | [ActionGround: Training-Free Runtime Refinement of Frozen VLA Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33256.pdf) | AgileX PiPER | 定性抓放；[证据](research/robot_hardware/evidence.md#paper-2609.33256) |
| <a id="paper-2609.33269"></a>2609.33269<br>WAM | [Q-WAM: 4-Bit Quantization of World Action Models with Action-Subspace Protection](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33269.pdf) | Unitree G1；UR3 ×2 | 量化评测；[证据](research/robot_hardware/evidence.md#paper-2609.33269) |
| <a id="paper-2609.33378"></a>2609.33378<br>VLA | [RECURSIVE HARNESS DISTILLATION ACROSS AGENTS FOR ROBOT MANIPULATION](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33378.pdf) | Franka Panda | 三任务；[证据](research/robot_hardware/evidence.md#paper-2609.33378) |
| <a id="paper-2609.33575"></a>2609.33575<br>VLA | [SLIP-VLA: Single-Step Latent Imagination for Policy Learning in Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33575.pdf) | AgileX Cobot Magic（PiPER＋TRACER） | 移动底盘存在不代表每项任务都用底盘；[证据](research/robot_hardware/evidence.md#paper-2609.33575) |
| <a id="paper-2609.33647"></a>2609.33647<br>VLA | [InfraVLA: Extending Vision-Language-Action Navigation with Infrastructure Cameras](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33647.pdf) | Boston Dynamics Spot | 基础设施相机导航；[证据](research/robot_hardware/evidence.md#paper-2609.33647) |
| <a id="paper-2609.33748"></a>2609.33748<br>WAM | [AnyStep-WAM: Budget-Aligned Distillation and Adaptive Inference for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33748.pdf) | Unitree G1D | 原文 G1D，不归并其他厂商G1；[证据](research/robot_hardware/evidence.md#paper-2609.33748) |
| <a id="paper-2609.33765"></a>2609.33765<br>VLA | [Principal Steering Subspaces for Online Adaptation of Frozen Generative Robot Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33765.pdf) | 型号未明 | 在线适配真实控制；[证据](research/robot_hardware/evidence.md#paper-2609.33765) |
| <a id="paper-2609.33832"></a>2609.33832<br>WAM | [Achieve What You Imagined: Learning to Align Actions with Visual Plans](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33832.pdf) | UR5 | 视觉计划动作对齐；[证据](research/robot_hardware/evidence.md#paper-2609.33832) |
| <a id="paper-2609.33944"></a>2609.33944<br>WAM | [ReSync: Re-Aligning the Two Clocks of Asynchronous World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.33944.pdf) | Mobile ALOHA（臂型号未明） | 配对时序实验；[证据](research/robot_hardware/evidence.md#paper-2609.33944) |
| <a id="paper-2609.34061"></a>2609.34061<br>VLA | [QUANTILE HEAD FOR VISION-LANGUAGE-ACTION MODELS](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34061.pdf) | AgileX PiPER | 两项任务；[证据](research/robot_hardware/evidence.md#paper-2609.34061) |
| <a id="paper-2609.34170"></a>2609.34170<br>VLA | [RAVEL: ASYNCHRONOUS ROLLING INFERENCE FOR FLOW-BASED VISION–LANGUAGE–ACTION MODELS](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34170.pdf) | xArm7 | 乒乓球；[证据](research/robot_hardware/evidence.md#paper-2609.34170) |
| <a id="paper-2609.34175"></a>2609.34175<br>VLA | [FailPatch: Failure Residual Patching for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34175.pdf) | 双6-DoF机械臂（型号未明） | 失败残差修补；[证据](research/robot_hardware/evidence.md#paper-2609.34175) |
| <a id="paper-2609.34199"></a>2609.34199<br>WAM | [WB-WAM: Heterogeneous Body-Hand Pre-training for Humanoid Loco-Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34199.pdf) | Unitree G1＋Wuji 手 ×2 | 全身移动操作；[证据](research/robot_hardware/evidence.md#paper-2609.34199) |
| <a id="paper-2609.34206"></a>2609.34206<br>VLA | [WorldGuide: Learning Success–Failure Boundaries in Latent World Models for Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34206.pdf) | ARX LIFT2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.34206) |
| <a id="paper-2609.34220"></a>2609.34220<br>VLA | [mmHRI: Towards Privacy-Preserving Human-Robot Interaction with Millimeter-Wave Radar](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34220.pdf) | Galaxea R1 Lite | 毫米波人机交互；[证据](research/robot_hardware/evidence.md#paper-2609.34220) |
| <a id="paper-2609.34250"></a>2609.34250<br>WAM | [WAM-OPD: Sharpening World Action Models via On-Policy Distillation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34250.pdf) | UR5e ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.34250) |
| <a id="paper-2609.34319"></a>2609.34319<br>VLA | [Text-Vision Synergistic Token Caching: A Training-Free Framework for Efficient Vision-Language-Action Inference](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34319.pdf) | Franka FR3 | token缓存；[证据](research/robot_hardware/evidence.md#paper-2609.34319) |
| <a id="paper-2609.34384"></a>2609.34384<br>VLA | [RoboIRGBench: Benchmarking Implicit Referential Grounding in Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34384.pdf) | Franka FR3 | 隐式指代测试；[证据](research/robot_hardware/evidence.md#paper-2609.34384) |
| <a id="paper-2609.34414"></a>2609.34414<br>WAM | [From World Models to World Action Models: Rethinking Next-State Prediction](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34414.pdf) | Fourier GR-1（待核实体型号） | 不能仅凭 Teleop-GR1 数据名确证；[证据](research/robot_hardware/evidence.md#paper-2609.34414) |
| <a id="paper-2609.34550"></a>2609.34550<br>VLA | [Gaze Prompts: Temporally Dense Human Attention for Vision–Language–Action Fine-Tuning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34550.pdf) | LingYu Robotics TeleAvatar | 移动双臂；眼动示教；[证据](research/robot_hardware/evidence.md#paper-2609.34550) |
| <a id="paper-2609.34688"></a>2609.34688<br>VLA | [Unified Trajectory Matching Policy Optimization: Diverse T2I Generation and VLA Generalization](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34688.pdf) | 机器人型号未明 | Uni-TMPO真实评估；[证据](research/robot_hardware/evidence.md#paper-2609.34688) |
| <a id="paper-2609.34792"></a>2609.34792<br>VLA | [D2-VLA: Dual-Memory Dual-Frequency Vision-Language-Action Model for Long Dynamic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34792.pdf) | Unitree G1 | 动态长程操作；[证据](research/robot_hardware/evidence.md#paper-2609.34792) |
| <a id="paper-2609.34893"></a>2609.34893<br>WAM | [ECHO: Event-Augmented Context with Hindsight and Outlook for Wrist-Only Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34893.pdf) | Astribot S1 | 腕部视角操作；[证据](research/robot_hardware/evidence.md#paper-2609.34893) |
| <a id="paper-2609.34911"></a>2609.34911<br>WAM/VLA | [Don't Throw Away the Tail: Action Upcycling for Policy Acceleration](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34911.pdf) | I2RT YAM | 八项任务424条示范；[证据](research/robot_hardware/evidence.md#paper-2609.34911) |
| <a id="paper-2609.34944"></a>2609.34944<br>VLA | [ADJOINT GUIDANCE FLOW: AMORTIZED CRITIC GUIDANCE FOR VLA POLICIES](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34944.pdf) | I2RT YAM | 6DoF；两相机两任务；[证据](research/robot_hardware/evidence.md#paper-2609.34944) |
| <a id="paper-2609.34968"></a>2609.34968<br>WAM/VLA | [RoboFL: Federated Expert Assembly for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34968.pdf) | Franka FR3 | Robotiq 2F-85；[证据](research/robot_hardware/evidence.md#paper-2609.34968) |
| <a id="paper-2609.34981"></a>2609.34981<br>WAM | [What Makes World Action Models Generalize? An Empirical Study of Test-Time Future Modeling](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34981.pdf) | AgileX ALOHA（臂型号未明） | 不是据名称推断Trossen ALOHA；[证据](research/robot_hardware/evidence.md#paper-2609.34981) |
| <a id="paper-2609.34982"></a>2609.34982<br>VLA | [ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34982.pdf) | RealMan Gen72 ×2 | 三台D435；三项双臂任务；[证据](research/robot_hardware/evidence.md#paper-2609.34982) |
| <a id="paper-2609.35003"></a>2609.35003<br>VLA | [LEARNING TO ACT UNDER VISUAL INTERRUPTIONS WITH VISION-LANGUAGE-ACTION MODELS](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.35003.pdf) | AgiBot G2 | 视觉中断；[证据](research/robot_hardware/evidence.md#paper-2609.35003) |
| <a id="paper-2609.35231"></a>2609.35231<br>VLA | [Zero-Shot Reactive Obstacle Avoidance for Generative Robot Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.35231.pdf) | UR5 | 动态障碍；[证据](research/robot_hardware/evidence.md#paper-2609.35231) |
| <a id="paper-2609.35249"></a>2609.35249<br>WAM/VLA | [Spatial Grafting: Grounding 3D Features for Flow-Matching Robot Policies](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.35249.pdf) | UR5e；AgileX PiPER ×2；Galaxea R1 Pro | 三平台；[证据](research/robot_hardware/evidence.md#paper-2609.35249) |
| <a id="paper-2609.35432"></a>2609.35432<br>VLA | [Self-Evolving Coding Agents: From Digital Programs to Physical-World Intelligence](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.35432.pdf) | AgileX 双臂（版本未明） | HexaAnything真实桌面任务；不将PhyBench算机器人型号；[证据](research/robot_hardware/evidence.md#paper-2609.35432) |
| <a id="paper-2609.35450"></a>2609.35450<br>VLA | [Uni-VLaT: Whole-Body Tactile Adaptation of VLA Policies for Humanoid Loco-Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.35450.pdf) | Unitree G1 | 全身自制织物触觉；[证据](research/robot_hardware/evidence.md#paper-2609.35450) |
| <a id="paper-2609.35469"></a>2609.35469<br>VLA | [Rethinking Causal Action Tokenization with Conditional Annealing in Flow Matching](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.35469.pdf) | Franka（版本未明） | 三类真实任务；[证据](research/robot_hardware/evidence.md#paper-2609.35469) |
| <a id="paper-2609.35575"></a>2609.35575<br>VLA | [F4R: Failure-Driven Recognition, Reconstruction, Refinement, and Redeployment for Continual Robot Self-Improvement](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.35575.pdf) | UR5e | Robotiq 2F-85；[证据](research/robot_hardware/evidence.md#paper-2609.35575) |
| <a id="paper-2609.35709"></a>2609.35709<br>VLA | [HUMANOID LOCO-MANIPULATION WITH DISCRETE VLA MODEL](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.35709.pdf) | Unitree G1-comp | Dex3 双手；[证据](research/robot_hardware/evidence.md#paper-2609.35709) |
| <a id="paper-2609.35761"></a>2609.35761<br>VLA | [DexRoam: Learning Mobile Bimanual Dexterous Manipulation from Egocentric Whole-Body Human Demonstrations](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.35761.pdf) | Astribot＋XHand ×2 | 移动灵巧操作；[证据](research/robot_hardware/evidence.md#paper-2609.35761) |
| <a id="paper-2609.36416"></a>2609.36416<br>VLA | [FineART: Fine-grained Annotated Robotic Trajectory Dataset and Vision-Language-Action Model for Bimanual Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.36416.pdf) | ALOHA（臂型号未明） | 六任务实体评估需核对协议；[证据](research/robot_hardware/evidence.md#paper-2609.36416) |
| <a id="paper-2609.36471"></a>2609.36471<br>WAM | [Staircase Policy: Streaming Inference for World-Action Models with Large Action Chunks](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.36471.pdf) | UFACTORY xArm 850 | 原文命名，是否等同UF850见官方/本文；[证据](research/robot_hardware/evidence.md#paper-2609.36471) |
| <a id="paper-2609.36588"></a>2609.36588<br>VLA | [COOPERATIVE MULTI-AGENT VISION-LANGUAGE-ACTION MODELS VIA REINFORCED FINE TUNING](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.36588.pdf) | Franka ×2（版本未明） | 多机器人协作；[证据](research/robot_hardware/evidence.md#paper-2609.36588) |
| <a id="paper-2609.36774"></a>2609.36774<br>VLA | [LEXICONVLA: LEARNING REUSABLE ATOMIC ACTION CODEBOOKS FOR UNSEEN TASKS](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.36774.pdf) | Franka FR3 | 新任务原子动作；[证据](research/robot_hardware/evidence.md#paper-2609.36774) |
| <a id="paper-2609.37150"></a>2609.37150<br>VLA | [CORE-VLA: PRESERVING CROSS-VIEW COORDINATION IN VLAS UNDER CAMERA SHIFTS](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.37150.pdf) | UR7e | 跨视角；[证据](research/robot_hardware/evidence.md#paper-2609.37150) |
| <a id="paper-2609.37165"></a>2609.37165<br>VLA | [Disentangling Spurious Correlations in Vision-Language-Action Models via Predicting Domain-Invariant Latent Lookahead](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.37165.pdf) | UR5e | Robotiq 两指；[证据](research/robot_hardware/evidence.md#paper-2609.37165) |
| <a id="paper-2609.37181"></a>2609.37181<br>VLA | [EGOHUMANOID-V2: HUMAN-TO-HUMANOID TRANSFER OF COORDINATED WHOLE-BODY SKILLS FOR LOCO-MANIPULATION](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.37181.pdf) | Unitree G1＋BrainCo Revo2 | 人到人形全身迁移；[证据](research/robot_hardware/evidence.md#paper-2609.37181) |
| <a id="paper-2609.37250"></a>2609.37250<br>WAM | [V-JEPA Policy: Building Effective World-Action Models on Predictive Visual Latents](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.37250.pdf) | Tianji Marvin | 四视角双臂；[证据](research/robot_hardware/evidence.md#paper-2609.37250) |
| <a id="paper-2609.37307"></a>2609.37307<br>VLA | [Remember What You Did: Action-History Memory with Dual-Expert Denoising for Long-Horizon Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.37307.pdf) | DOBOT Nova（版本未明） | 四项记忆任务；未细分Nova2/5；[证据](research/robot_hardware/evidence.md#paper-2609.37307) |
| <a id="paper-2609.37334"></a>2609.37334<br>VLA | [TAMING VLAS UNDER ROBOT EXECUTION ERRORS: SELF-COMPENSATION AND STRESS TESTING](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.37334.pdf) | AgileX PiPER ×2 | 执行误差；[证据](research/robot_hardware/evidence.md#paper-2609.37334) |
| <a id="paper-2609.37398"></a>2609.37398<br>WAM | [Direct Experience World-Model Optimization: Learning the World Beyond Action Imitation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.37398.pdf) | Xingchen＋Wuji手；Tianji＋Sharpa手 | 附录C.1；两种本体；[证据](research/robot_hardware/evidence.md#paper-2609.37398) |
| <a id="paper-2609.37530"></a>2609.37530<br>VLA | [RAWVLA: EMBODIED NEURAL IMAGE SIGNAL PROCESSOR FOR ROBOTIC MANIPULATION](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.37530.pdf) | ROKAE AR5 ×2 | Songling夹爪；HIKROBOT相机；[证据](research/robot_hardware/evidence.md#paper-2609.37530) |
| <a id="paper-2609.37721"></a>2609.37721<br>WAM | [CogWAM: Aligning Semantic Cognition with World Action Modeling via Event-Driven Interfaces](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.37721.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.37721) |
| <a id="paper-2609.37771"></a>2609.37771<br>VLA | [FASTER AND BETTER? BENCHMARK BUGS AND DESIGN LIMITATIONS DISTORT THE EVALUATION OF VISION-LANGUAGE-ACTION ACCELERATION](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.37771.pdf) | Mobile ALOHA（臂型号未明） | 加速评测审计；[证据](research/robot_hardware/evidence.md#paper-2609.37771) |
| <a id="paper-2609.37772"></a>2609.37772<br>VLA | [URGENT ACTIONS GO FIRST: URGENCY-AWARE DENOISING FOR REAL-TIME VLA CONTROL](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.37772.pdf) | Dobot Nova2 | 实时去噪；[证据](research/robot_hardware/evidence.md#paper-2609.37772) |
| <a id="paper-2609.37793"></a>2609.37793<br>WAM | [MVG-WAM: Multiple View Geometry-Aware World-Action Modeling for Robotic Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.37793.pdf) | AgileX Cobot Magic | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.37793) |
| <a id="paper-2609.37810"></a>2609.37810<br>WAM/VLA | [Explore, Execute, Evolve: A Skill Acquisition and Reuse Loop for Embodied Agents](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.37810.pdf) | AgileX PiPER（右臂） | 技能积累；触觉；[证据](research/robot_hardware/evidence.md#paper-2609.37810) |
| <a id="paper-2609.38057"></a>2609.38057<br>WAM | [EVO-WAM: Evolving World Action Models through Video-Action Verification](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.38057.pdf) | Franka（版本未明） | 三项长程任务；[证据](research/robot_hardware/evidence.md#paper-2609.38057) |
| <a id="paper-2609.38078"></a>2609.38078<br>VLA | [MotorMind: Scaffolding General Vision Language Models for Zero-Shot Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.38078.pdf) | xArm6 | 零样本大模型控制；[证据](research/robot_hardware/evidence.md#paper-2609.38078) |
| <a id="paper-2609.38163"></a>2609.38163<br>WAM | [Rethinking Representations for World-Action Modeling](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.38163.pdf) | AgileX PiPER | 表征对比；[证据](research/robot_hardware/evidence.md#paper-2609.38163) |
| <a id="paper-2609.38164"></a>2609.38164<br>VLA | [Rho: A Foundation for Efficiently Adaptable VLA Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.38164.pdf) | Franka FR3 Duo；UR AI Trainer；YAM Box | 三种实体双臂系统；[证据](research/robot_hardware/evidence.md#paper-2609.38164) |
| <a id="paper-2609.38984"></a>2609.38984<br>WAM | [Sparse-WAM: Accelerating World Action Models via Action-Guided Sparse Imagination](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.38984.pdf) | AgileX Cobot Magic | 稀疏想象；[证据](research/robot_hardware/evidence.md#paper-2609.38984) |
| <a id="paper-2609.39056"></a>2609.39056<br>WAM | [SteerQuant: Steering Quantization Error with Action-Guided Scaling in World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.39056.pdf) | AgileX Cobot Magic | 动作引导量化；[证据](research/robot_hardware/evidence.md#paper-2609.39056) |
| <a id="paper-2609.39388"></a>2609.39388<br>WAM | [UniWAM Technical Report: Unified Mobile Manipulation via Mixed-Stream World-Action Modeling and Manipulation Anchor Pose Supervision](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.39388.pdf) | AgileX Cobot Magic；AgiBot G2；Franka FR3 ×2；AgileX 单臂（版本未明） | 导航与移动操作分平台评估；[证据](research/robot_hardware/evidence.md#paper-2609.39388) |
| <a id="paper-2609.39763"></a>2609.39763<br>WAM | [DiffWAM: A Fast and Efficient Navigation World Action Model](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.39763.pdf) | NCS-α-pro；NCS-β；NCS-Thor-preview | 无人机定性飞行；[证据](research/robot_hardware/evidence.md#paper-2609.39763) |
| <a id="paper-2609.39870"></a>2609.39870<br>WAM | [Magic-W0: A Structured World-Action Foundation Model for Physical Intelligence](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.39870.pdf) | 型号未明 | 五类真实任务；[证据](research/robot_hardware/evidence.md#paper-2609.39870) |
| <a id="paper-2609.39873"></a>2609.39873<br>WAM | [SplineWAM: Adaptive Action Horizons for World Action Models via B-Spline Representations](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.39873.pdf) | ARX AC-One；ARX Lift-2S | 固定/移动基座；[证据](research/robot_hardware/evidence.md#paper-2609.39873) |
| <a id="paper-2609.39973"></a>2609.39973<br>WAM | [EWAM: Emergent Depth-Wise Specialization in a Unified Embodied Model -- From Semantic Understanding through Visual Foresight to Action](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.39973.pdf) | Franka（单臂）；Dobot（双臂）；Unitree G1-D | 三平台；臂具体代际未明；[证据](research/robot_hardware/evidence.md#paper-2609.39973) |
| <a id="paper-2609.40219"></a>2609.40219<br>WAM | [Learning Skills from Historical Action Trajectories: Action Experience Dictionary for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.40219.pdf) | Spirit AI MOZ1；ROKAE AR5-5 0.7 | 附录E.2细化AR5型号；[证据](research/robot_hardware/evidence.md#paper-2609.40219) |
| <a id="paper-2609.40341"></a>2609.40341<br>WAM | [Ego4WAM: What Matters When Scaling Egocentric Human Data for Robot Learning?](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.40341.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2609.40341) |
| <a id="paper-2610.00638"></a>2610.00638<br>WAM | [TacDyn-WAM: Learning Implicit Tactile Dynamics in a Heterogeneous Visuo-Tactile World Action Model](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.00638.pdf) | Franka FR3 | Xense 双触觉；[证据](research/robot_hardware/evidence.md#paper-2610.00638) |
| <a id="paper-2610.00899"></a>2610.00899<br>VLA | [TOAST: Stochastic Robot Action Tokenization for Autoregressive Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.00899.pdf) | Franka FR3 | Robotiq 2F-85；[证据](research/robot_hardware/evidence.md#paper-2610.00899) |
| <a id="paper-2610.00913"></a>2610.00913<br>VLA | [eRLT: Efficient VLA Reinforcement Learning via Action-Relevant Token Routing](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.00913.pdf) | Rainbow Robotics RB-Y1 | USB和排线插入在线强化学习；[证据](research/robot_hardware/evidence.md#paper-2610.00913) |
| <a id="paper-2610.00981"></a>2610.00981<br>VLA | [NarrativeFlow: Flow-Based Vision-Language-Action Model Using Robot Velocity Fields](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.00981.pdf) | Toyota HSR | Human Support Robot；[证据](research/robot_hardware/evidence.md#paper-2610.00981) |
| <a id="paper-2610.00982"></a>2610.00982<br>VLA | [Divide-and-Remember: Recursive Action-Relevant Memory for Long-Horizon VLA Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.00982.pdf) | Franka Emika Panda＋Robotiq 2F-140 | 附录图26；[证据](research/robot_hardware/evidence.md#paper-2610.00982) |
| <a id="paper-2610.01698"></a>2610.01698<br>WAM | [ActiveWAM: Evidence-Aware Active Vision for World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.01698.pdf) | AIRBOT Play ×2 | D455＋IQR云台；段间人工衔接；[证据](research/robot_hardware/evidence.md#paper-2610.01698) |
| <a id="paper-2610.01741"></a>2610.01741<br>VLA | [ATI-VLA: Action-Centric Predictive Vision-Language-Action Models via Actionable Alignment Then Adaptive Injection](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.01741.pdf) | Galaxea R1 Lite；AgileX Cobot Magic | 长程任务；[证据](research/robot_hardware/evidence.md#paper-2610.01741) |
| <a id="paper-2610.01856"></a>2610.01856<br>VLA | [ChunkVLA-AM: Parallel Action Chunking for Vision-Language-Action Robot Control in Additive Manufacturing](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.01856.pdf) | Franka FR3 | 42次真机试验；主机并行推理；[证据](research/robot_hardware/evidence.md#paper-2610.01856) |
| <a id="paper-2610.02054"></a>2610.02054<br>WAM/VLA | [UniWAM: Unified World-Action Model](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02054.pdf) | AgileX PiPER ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2610.02054) |
| <a id="paper-2610.02120"></a>2610.02120<br>WAM | [SkeleWAM: Skeleton World-Action Modeling for Efficient Robotic Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02120.pdf) | ARX R5 | 骨架建模；[证据](research/robot_hardware/evidence.md#paper-2610.02120) |
| <a id="paper-2610.02323"></a>2610.02323<br>VLA | [World-Calibrated Proposal-to-Action Flow for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02323.pdf) | 型号未明 | 提案动作流；[证据](research/robot_hardware/evidence.md#paper-2610.02323) |
| <a id="paper-2610.02360"></a>2610.02360<br>VLA | [SocialVLA: A Social Perception Gateway for Human-Reaction-Based Failure Detection and Recovery in VLA Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02360.pdf) | Unitree G1 | 人类反应失败检测；[证据](research/robot_hardware/evidence.md#paper-2610.02360) |
| <a id="paper-2610.02368"></a>2610.02368<br>WAM | [Rethinking World-Action Model for Compositional and In-Context Robotic Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02368.pdf) | AgiBot A2 | 组合任务；[证据](research/robot_hardware/evidence.md#paper-2610.02368) |
| <a id="paper-2610.02398"></a>2610.02398<br>WAM | [Keep the Effect, Drop the Actor: Programmable Effect-to-Execution World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02398.pdf) | Franka FR3 | 附录L；26Hz操作空间控制；[证据](research/robot_hardware/evidence.md#paper-2610.02398) |
| <a id="paper-2610.02508"></a>2610.02508<br>WAM | [World Action Modeling with Progressive Visual Planning](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02508.pdf) | Franka FR3 | 沿用DROID配置不代表Panda；[证据](research/robot_hardware/evidence.md#paper-2610.02508) |
| <a id="paper-2610.02626"></a>2610.02626<br>VLA | [Imagine the Future, Internalize the Gist: Efficient VLA Reasoning via Internalized Spatiotemporal Imagination](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02626.pdf) | UR3 | 螺丝刀放置；[证据](research/robot_hardware/evidence.md#paper-2610.02626) |
| <a id="paper-2610.02717"></a>2610.02717<br>VLA | [RoboBridge: A Self-Evolving Embodied Agent Framework for Sim-to-Real Transfer](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02717.pdf) | 型号未明 | sim2real技能演化；[证据](research/robot_hardware/evidence.md#paper-2610.02717) |
| <a id="paper-2610.02784"></a>2610.02784<br>VLA | [SimpleTouch: Can Vision-Language-Action Models Master Contact-Rich Manipulation Without Tactile Policy Pretraining?](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02784.pdf) | AgileX PiPER | GelSight Mini；[证据](research/robot_hardware/evidence.md#paper-2610.02784) |
| <a id="paper-2610.02804"></a>2610.02804<br>VLA | [SARI: Phase-Split Sim-Real Co-Training for Contact-Rich Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02804.pdf) | Franka FR3 | Franka Hand；[证据](research/robot_hardware/evidence.md#paper-2610.02804) |
| <a id="paper-2610.02832"></a>2610.02832<br>WAM/VLA | [FastOPD: On-Policy Distillation for Lightweight VLA Deployment](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02832.pdf) | I2RT YAM（右臂） | 7维动作含夹爪；[证据](research/robot_hardware/evidence.md#paper-2610.02832) |
| <a id="paper-2610.02840"></a>2610.02840<br>WAM | [PointWAM: 3D World Action Modeling for Dexterous Robotic Manipulation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02840.pdf) | OpenArm（双7-DoF）＋Inspire RH56F1 ×2 | 原文称 humanoid；[证据](research/robot_hardware/evidence.md#paper-2610.02840) |
| <a id="paper-2610.02898"></a>2610.02898<br>VLA | [MixVLA: Adaptive Mixing of Non-Invariant Information for Generalizable Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.02898.pdf) | Franka（版本未明） | 照明变化；[证据](research/robot_hardware/evidence.md#paper-2610.02898) |
| <a id="paper-2610.03079"></a>2610.03079<br>WAM | [RIFAR: Reliability and Forgetting-Aware Replay for Continual Robot Learning](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.03079.pdf) | Franka（版本未明） | 三任务持续学习；[证据](research/robot_hardware/evidence.md#paper-2610.03079) |
| <a id="paper-2610.03391"></a>2610.03391<br>WAM | [Native Action-Prior Learning from Videos for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.03391.pdf) | Franka FR3 | DROID后训练零样本；[证据](research/robot_hardware/evidence.md#paper-2610.03391) |
| <a id="paper-2610.03476"></a>2610.03476<br>VLA | [MobiAgent: Dual-Loop Recursive Policy Self-Improvement for Long-Horizon Mobile Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.03476.pdf) | Astribot S1 | 长程；[证据](research/robot_hardware/evidence.md#paper-2610.03476) |
| <a id="paper-2610.03959"></a>2610.03959<br>WAM | [LatentQuant: Preserving the Policy-Facing Latent Contract under NVFP4 VAE Quantization](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.03959.pdf) | XSquare Robot（版本未明） | 包含物理观测回放和交互式真机评估；[证据](research/robot_hardware/evidence.md#paper-2610.03959) |
| <a id="paper-2610.04009"></a>2610.04009<br>WAM/VLA | [SUAVE: Unified Video-Action Models via Masked Diffusion](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.04009.pdf) | xArm7 | DROID只是预训练数据；[证据](research/robot_hardware/evidence.md#paper-2610.04009) |
| <a id="paper-2610.04255"></a>2610.04255<br>VLA | [Grounded in Time: A Multi-Source Dataset and Benchmark for Temporal Grounding in Robotic Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.04255.pdf) | AgileX PiPER-X ×2 | 时间接地数据/闭环需分开；[证据](research/robot_hardware/evidence.md#paper-2610.04255) |
| <a id="paper-2610.04391"></a>2610.04391<br>VLA | [AgenticTactileVLA: Contact-Guided Execution-Time Supervision for Generalizable Dexterous Manipulation without VLA Retraining](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.04391.pdf) | Unitree G1＋BrainCo Revo2 | 触觉执行监督；[证据](research/robot_hardware/evidence.md#paper-2610.04391) |
| <a id="paper-2610.04607"></a>2610.04607<br>VLA | [ForeAct3D: Policy-Grounded Future World Modeling for VLA Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.04607.pdf) | AgileX（版本未明） | 三项几何任务；[证据](research/robot_hardware/evidence.md#paper-2610.04607) |
| <a id="paper-2610.04616"></a>2610.04616<br>VLA | [PerturBot: Breaking Shortcut Priors in Vision-Language-Action Models with Perturbative Training](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.04616.pdf) | AgileX PiPER-X ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2610.04616) |
| <a id="paper-2610.04681"></a>2610.04681<br>WAM/VLA | [RoboIRS: Inference-Time Internal Representation Steering for Generalist Robot Policies](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.04681.pdf) | 型号未明 | H1/H2 是任务编号，不能当人形型号；[证据](research/robot_hardware/evidence.md#paper-2610.04681) |
| <a id="paper-2610.04805"></a>2610.04805<br>VLA | [ExStereo: Lifting 2D Vision-Language-Action Models to 3D with Explicit Stereo Representations](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.04805.pdf) | AgileX PiPER ×2 | 双目；[证据](research/robot_hardware/evidence.md#paper-2610.04805) |
| <a id="paper-2610.04913"></a>2610.04913<br>WAM | [$R^2$-WAM: Repair-and-Reject Post-Training for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.04913.pdf) | AgileX PiPER ×2 | 修复拒绝后训练；[证据](research/robot_hardware/evidence.md#paper-2610.04913) |
| <a id="paper-2610.04933"></a>2610.04933<br>VLA | [DiVeR: Decision-Critical Verifier Learning for VLA Test-Time Scaling](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.04933.pdf) | Franka FR3 | 验证器；[证据](research/robot_hardware/evidence.md#paper-2610.04933) |
| <a id="paper-2610.05025"></a>2610.05025<br>VLA | [Triggering Generalist Reasoning via Predictive Uncertainty for Dual-System VLA](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.05025.pdf) | SO-101 | 双系统触发；[证据](research/robot_hardware/evidence.md#paper-2610.05025) |
| <a id="paper-2610.05026"></a>2610.05026<br>VLA | [GeoBridge-VLA: Geometry-Aware Residual Adaptation for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.05026.pdf) | ROBOTIS OMY | 四任务；[证据](research/robot_hardware/evidence.md#paper-2610.05026) |
| <a id="paper-2610.05230"></a>2610.05230<br>VLA | [Vela: Scaling Vision-Language-Action Models with Adaptive Action Curve Parametrization](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.05230.pdf) | 型号未明 | 蛋饼烹饪；[证据](research/robot_hardware/evidence.md#paper-2610.05230) |
| <a id="paper-2610.05273"></a>2610.05273<br>VLA | [When and What to Prune? Stage-Aware Visual Token Pruning for Efficient VLA](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.05273.pdf) | AgileX PiPER | 视觉剪枝；[证据](research/robot_hardware/evidence.md#paper-2610.05273) |
| <a id="paper-2610.05418"></a>2610.05418<br>VLA | [EvoMem-VLA: State-Evolution Memory for Long-Horizon Robot Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.05418.pdf) | Franka（版本未明）；AgileX Cobot | 双本体记忆；[证据](research/robot_hardware/evidence.md#paper-2610.05418) |
| <a id="paper-2610.05483"></a>2610.05483<br>WAM | [FLEX-WAM: Flexible Block-Causal World-Action Models for Long-Horizon Imagination and Planning](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.05483.pdf) | OpenArm V1 ×2 | 闭环play策略及模型现实偏差检测；Unitree G1仅离线数据滚动预测，不计G1部署；[证据](research/robot_hardware/evidence.md#paper-2610.05483) |
| <a id="paper-2610.05719"></a>2610.05719<br>VLA | [When to Switch: Reliable Action-Chunk Extension for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.05719.pdf) | AgileX PiPER | chunk延长；[证据](research/robot_hardware/evidence.md#paper-2610.05719) |
| <a id="paper-2610.05755"></a>2610.05755<br>VLA | [Demonstration-Calibrated Port-Hamiltonian Retuning for Manipulation Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.05755.pdf) | Franka FR3 | 控制器调谐；[证据](research/robot_hardware/evidence.md#paper-2610.05755) |
| <a id="paper-2610.06184"></a>2610.06184<br>VLA | [Arm-wise Compositional Generalization in Dual-Arm Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.06184.pdf) | SO-101 | 双臂组合泛化；[证据](research/robot_hardware/evidence.md#paper-2610.06184) |
| <a id="paper-2610.06235"></a>2610.06235<br>WAM/VLA | [Encoded but Not in Control: Revealing the Grounding Gap in Vision-Language Robot Policies](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.06235.pdf) | 型号未明 | 指代接地；[证据](research/robot_hardware/evidence.md#paper-2610.06235) |
| <a id="paper-2610.06617"></a>2610.06617<br>WAM | [RealtimeWAM: One-Step Asynchronous World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.06617.pdf) | AgileX PiPER ×2 | 定性T恤折叠；[证据](research/robot_hardware/evidence.md#paper-2610.06617) |
| <a id="paper-2610.06926"></a>2610.06926<br>VLA | [SWAP: Stepwise Action Policy Routing for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.06926.pdf) | Franka Panda | DROID真实路由；[证据](research/robot_hardware/evidence.md#paper-2610.06926) |
| <a id="paper-2610.06965"></a>2610.06965<br>WAM | [ACG-WAM: World-Action Modeling via Action-Conditioned Geometric Latent Prediction](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.06965.pdf) | LimX TRON2＋Wuji 手 | 三项操作；[证据](research/robot_hardware/evidence.md#paper-2610.06965) |
| <a id="paper-2610.07558"></a>2610.07558<br>VLA | [Seeing the Invisible: Physics-Guided Visual Prompting for Temperature- and Radiation-Aware VLA Navigation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.07558.pdf) | Unitree Go2 EDU | 温度/辐射导航；[证据](research/robot_hardware/evidence.md#paper-2610.07558) |
| <a id="paper-2610.07652"></a>2610.07652<br>VLA | [SMART: Zero-Shot Sim-to-Real Articulated Object Manipulation via Large-Scale Synthetic Pretraining](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.07652.pdf) | RealMan RM75；ARX AC1；Galaxea R1 Pro | 铰接物体零样本；[证据](research/robot_hardware/evidence.md#paper-2610.07652) |
| <a id="paper-2610.07696"></a>2610.07696<br>VLA | [ESP: Energy-Score Policy for One-Step Multimodal Action Generation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.07696.pdf) | Franka（版本未明） | 单步策略；[证据](research/robot_hardware/evidence.md#paper-2610.07696) |
| <a id="paper-2610.07756"></a>2610.07756<br>VLA | [StairVLA: Stage-Aware Hierarchical Action Generation for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.07756.pdf) | AgileX PiPER-X | 阶段动作；[证据](research/robot_hardware/evidence.md#paper-2610.07756) |
| <a id="paper-2610.07922"></a>2610.07922<br>WAM | [OpenWAM: An Open Framework for Composable World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.07922.pdf) | Franka FR3 ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2610.07922) |
| <a id="paper-2610.07946"></a>2610.07946<br>VLA | [Adapting Vision-Language-Action Models to Unknown Visual Disruptions During Execution](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.07946.pdf) | SO-101 | 数字与物理视觉扰动；[证据](research/robot_hardware/evidence.md#paper-2610.07946) |
| <a id="paper-2610.08133"></a>2610.08133<br>VLA | [VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.08133.pdf) | AgileX PiPER-X | 双臂；[证据](research/robot_hardware/evidence.md#paper-2610.08133) |
| <a id="paper-2610.08150"></a>2610.08150<br>VLA | [ViDAL: A Visual Dynamics-Grounded Action Latent Space for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.08150.pdf) | Franka（版本未明）；ARX（版本未明） | 双平台；[证据](research/robot_hardware/evidence.md#paper-2610.08150) |
| <a id="paper-2610.08425"></a>2610.08425<br>VLA | [MIM-VLA: Learning Physical Interaction Representations from Gripper Motor Feedback](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.08425.pdf) | OMY F3M | 6DoF从臂；Dynamixel双指夹爪；电流反馈不是标定力；[证据](research/robot_hardware/evidence.md#paper-2610.08425) |
| <a id="paper-2610.08640"></a>2610.08640<br>WAM | [RIWANav: Recursive World-Action Models with Self-Improvement for Urban Navigation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.08640.pdf) | 移动平台（型号未明） | 真实闭环导航；[证据](research/robot_hardware/evidence.md#paper-2610.08640) |
| <a id="paper-2610.09016"></a>2610.09016<br>VLA | [PAIR: Bridging Perception and Action in Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.09016.pdf) | UFACTORY UF850 | 七项任务293条真实示范；[证据](research/robot_hardware/evidence.md#paper-2610.09016) |
| <a id="paper-2610.09144"></a>2610.09144<br>VLA | [DIVA: Dual-Space Intent-Aware Visual Attenuation for Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.09144.pdf) | UFACTORY UF850 | OOD视觉扰动；[证据](research/robot_hardware/evidence.md#paper-2610.09144) |
| <a id="paper-2610.09170"></a>2610.09170<br>VLA | [Beyond Reconstruction: What Matters in Action Tokenization for Robot Policies?](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.09170.pdf) | UR5 | 动作token对比；[证据](research/robot_hardware/evidence.md#paper-2610.09170) |
| <a id="paper-2610.09228"></a>2610.09228<br>VLA | [Co-Evolving Robot Orchestrators and Policies through Deployment](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.09228.pdf) | DROID 平台（具体型号待核） | 策略与编排联合演化；[证据](research/robot_hardware/evidence.md#paper-2610.09228) |
| <a id="paper-2610.09427"></a>2610.09427<br>WAM | [Event-Aligned Visual Action Reasoning for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.09427.pdf) | Unitree G1＋Dex1-1夹爪 | 附录C.1；三项真实任务；[证据](research/robot_hardware/evidence.md#paper-2610.09427) |
| <a id="paper-2610.09451"></a>2610.09451<br>VLA | [TempoBridge: Language-Guided Tempo Control for Vision-Language-Action Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.09451.pdf) | xArm6 | 节奏控制；[证据](research/robot_hardware/evidence.md#paper-2610.09451) |
| <a id="paper-2610.09496"></a>2610.09496<br>VLA | [Sparse Feature Policy Unlearning Mitigates State Hallucination in Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.09496.pdf) | Franka FR3 | 状态幻觉；[证据](research/robot_hardware/evidence.md#paper-2610.09496) |
| <a id="paper-2610.09696"></a>2610.09696<br>VLA | [RoboPace: Contact-Aware Time-Optimal Retiming for Action-Chunk Policies](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.09696.pdf) | OpenArm v1.0 ×2＋YUBI 手 | 接触重定时；[证据](research/robot_hardware/evidence.md#paper-2610.09696) |
| <a id="paper-2610.09718"></a>2610.09718<br>VLA | [YUBI-STAG: Contact and Semantic-Rich Alignment for VLAs via Automated Video-Language Grounding](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.09718.pdf) | Franka（版本未明）＋YUBI 末端 | 小零件；[证据](research/robot_hardware/evidence.md#paper-2610.09718) |
| <a id="paper-2610.09734"></a>2610.09734<br>WAM | [ΔWAM: Distilling Action Tangent Fields into World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.09734.pdf) | AgileX NERO；Tianji（版本未明） | 插管、叠碗及烹饪；[证据](research/robot_hardware/evidence.md#paper-2610.09734) |
| <a id="paper-2610.09940"></a>2610.09940<br>VLA | [Juno: Taming Predictive Latents for Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.09940.pdf) | AgileX Cobot Magic（右臂） | 单臂真实任务；[证据](research/robot_hardware/evidence.md#paper-2610.09940) |
| <a id="paper-2610.09943"></a>2610.09943<br>VLA | [Many Ways to Succeed: Diversity-Driven RL Fine-Tuning for VLA Generalization](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.09943.pdf) | AgileX PiPER-X ×2 | 双臂；[证据](research/robot_hardware/evidence.md#paper-2610.09943) |
| <a id="paper-2610.10079"></a>2610.10079<br>WAM | [RealtimeWAM: How Fast Can I Run My World Action Model?](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.10079.pdf) | Franka FR3 | 传送带及动态任务；[证据](research/robot_hardware/evidence.md#paper-2610.10079) |
| <a id="paper-2610.10270"></a>2610.10270<br>WAM | [Video Prediction Policy 2: Predict Better, Act Better](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.10270.pdf) | ALOHA（臂型号未明） | 十类真实任务；[证据](research/robot_hardware/evidence.md#paper-2610.10270) |
| <a id="paper-2610.10384"></a>2610.10384<br>VLA | [OpenViTac: Learning and Benchmarking Visuo-Tactile Policies in a Unified Sim-and-Real Framework](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.10384.pdf) | 机械臂型号未明；Franka Hand／Robotiq夹爪 | GelSight／Xense／NeoteAI；夹爪品牌不能推出机械臂型号；[证据](research/robot_hardware/evidence.md#paper-2610.10384) |
| <a id="paper-2610.10437"></a>2610.10437<br>VLA | [Q-Learning with Scalar Adjoint Matching](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.10437.pdf) | Rainbow Robotics RB-Y1＋Wuji Hand 2 ×2 | 强化学习；[证据](research/robot_hardware/evidence.md#paper-2610.10437) |
| <a id="paper-2610.10528"></a>2610.10528<br>WAM | [Long-WAM: Scaling the Context of World-Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.10528.pdf) | Unitree G1；YAM | 动态及长程；[证据](research/robot_hardware/evidence.md#paper-2610.10528) |

## 真实数据／硬件测量与未决项（D／U）

以下不混入主要真机统计。尤其是数据中的手术机器人、驾驶车辆、训练语料平台，不等于本文新部署了对应策略。

| ID／目录 | 论文（本地原文） | 真机型号／配置 | 用途与辨析 |
|---|---|---|---|
| <a id="paper-2608.00391"></a>2608.00391<br>VLA · D | [The Gate, Not the Cache: Gate Provenance Bounds the Closed-Loop Reliability of Training-Free VLA Token Skipping](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.00391.pdf) | 机器人型号未明 | 真机平台推理时延测量；不按已确认策略执行统计；[证据](research/robot_hardware/evidence.md#paper-2608.00391) |
| <a id="paper-2608.03244"></a>2608.03244<br>WAM · D | [UniNav: A Unified World-Action Diffusion Model for Visual Navigation](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.03244.pdf) | 移动机器人（型号未明） | 明确真实采集数据；未确认实机闭环；[证据](research/robot_hardware/evidence.md#paper-2608.03244) |
| <a id="paper-2608.11204"></a>2608.11204<br>WAM · D | [Surgical WAM: A World-Action Model for Data-Efficient Surgical Robot Learning](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.11204.pdf) | da Vinci / JIGSAWS 数据 | 只用手术数据；没有实体 dVRK 闭环；[证据](research/robot_hardware/evidence.md#paper-2608.11204) |
| <a id="paper-2608.11769"></a>2608.11769<br>VLA · D | [Policy-Induced Hand Priors in Humanoid Dual-Arm Manipulation: Diagnosing and Mitigating Initial-Pose Dependence](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.11769.pdf) | Unitree G1（真实混训数据）；评测为仿真 | 原文明确在仿真干预初始关节姿态；不计G1实体策略部署；[证据](research/robot_hardware/evidence.md#paper-2608.11769) |
| <a id="paper-2608.20430"></a>2608.20430<br>WAM · U | [RISE: Adaptive Imagination for World Action Models](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2608.20430.pdf) | COWARobot（型号及实车性质待核） | 需区分离线驾驶日志；[证据](research/robot_hardware/evidence.md#paper-2608.20430) |
| <a id="paper-2609.16705"></a>2609.16705<br>VLA · D | [The Robot Data Factory](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.16705.pdf) | Unitree B2＋Z1；其他平台见原文 | 数据工厂及硬件采集；非单一 VLA 部署论文；[证据](research/robot_hardware/evidence.md#paper-2609.16705) |
| <a id="paper-2609.23580"></a>2609.23580<br>VLA · U | [TaskAnchor: Grounding Task State in Reactive VLAs for Long-Horizon Manipulation](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.23580.pdf) | 型号未明 | 摘要声称真机增益；正文证据有限；[证据](research/robot_hardware/evidence.md#paper-2609.23580) |
| <a id="paper-2609.24048"></a>2609.24048<br>WAM · D | [What Matters in Designing World Action Models: An Empirical Study](wam%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.24048.pdf) | 型号未明 | 真实轨迹离线诊断；不视为新闭环；[证据](research/robot_hardware/evidence.md#paper-2609.24048) |
| <a id="paper-2609.31313"></a>2609.31313<br>VLA · U | [Towards VLA-Dreamer: Refining VLA Behavior Using World Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.31313.pdf) | NICOL | 目标本体不等于已完成实体部署；[证据](research/robot_hardware/evidence.md#paper-2609.31313) |
| <a id="paper-2609.34794"></a>2609.34794<br>VLA · D | [WHERE DO EMBODIED DECISIONS COME FROM? RETHINKING LATENT AND EXPLICIT REASONING](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.34794.pdf) | Li Auto车辆数据（具体车型未明） | 真实驾驶数据上的离线决策与路径生成；[证据](research/robot_hardware/evidence.md#paper-2609.34794) |
| <a id="paper-2609.35039"></a>2609.35039<br>VLA · U | [SeeReach-VLA: Sign-Aware Edge Deployment for Vineyard Harvesting under Occlusion](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2609.35039.pdf) | 四足＋机械臂（型号未明） | 文末Results to be reported in the final version；不计已完成真机实验；[证据](research/robot_hardware/evidence.md#paper-2609.35039) |
| <a id="paper-2610.06598"></a>2610.06598<br>VLA · D | [SimForcing: Distilling Simulation Motion Priors into Real-Domain Robot World Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.06598.pdf) | Bridge／InternData-A1真实数据平台（本篇未确认新部署型号） | 真实域视频预测；不能据real-world predictions认定闭环真机；[证据](research/robot_hardware/evidence.md#paper-2610.06598) |
| <a id="paper-2610.10526"></a>2610.10526<br>VLA · D | [Rephrase Before You Act: Characterizing and Mitigating Language Sensitivity in Vision-Language-Action Models](vla%E8%AE%BA%E6%96%87/reading_reports/arxiv_cache/2610.10526.pdf) | Bridge／WidowX数据与仿真 | 12个sealed任务在SIMPLER中；另用真实专家轨迹，不是12项物理指令测试；[证据](research/robot_hardware/evidence.md#paper-2610.10526) |

## 维护方式和剩余边界

新增论文时，以 arXiv ID 更新 `research/robot_hardware/papers.json`；逐篇填写目录归属、状态、原文支持的硬件名称和实验角色。运行 `python3 research/robot_hardware/build_readme.py` 重建本页、全量审计表和摘录。`annotations.tsv` 保留本轮候选人工标注的中间记录，最终发布数据以 `papers.json` 为准。

本页的边界是：型号未披露、仅在图片中标注或检索无法可靠分辨的记录仍标为未明；U 不当成肯定的真机结果；N 不当成肯定的无真机结论。要做采购或复现实验，应先从对应 PDF 的 hardware/setup 附录确认代际、末端、相机和控制接口，本表不替代这些配置细节。