1M+ trajectories from 22 robot types pooled by 30+ labs; RT-X models trained on it beat single-robot baselines.30 多个实验室汇集 22 种机器人的 100 多万条轨迹；在其上训练的 RT-X 优于单机器人基线。
@misc{collaboration2023open,
title = {{Open X-Embodiment: Robotic Learning Datasets and RT-X Models}},
author = {Open X-Embodiment Collaboration and Abby O'Neill and Abdul Rehman and Abhinav Gupta and Abhiram Maddukuri and Abhishek Gupta and Abhishek Padalkar and Abraham Lee and Acorn Pooley and Agrim Gupta and Ajay Mandlekar and Ajinkya Jain and Albert Tung and Alex Bewley and Alex Herzog and Alex Irpan and Alexander Khazatsky and Anant Rai and Anchit Gupta and Andrew Wang and Andrey Kolobov and Anikait Singh and Animesh Garg and Aniruddha Kembhavi and Annie Xie and Anthony Brohan and Antonin Raffin and Archit Sharma and Arefeh Yavary and Arhan Jain and Ashwin Balakrishna and Ayzaan Wahid and Ben Burgess-Limerick and Beomjoon Kim and Bernhard Schölkopf and Blake Wulfe and Brian Ichter and Cewu Lu and Charles Xu and Charlotte Le and Chelsea Finn and Chen Wang and Chenfeng Xu and Cheng Chi and Chenguang Huang and Christine Chan and Christopher Agia and Chuer Pan and Chuyuan Fu and Coline Devin and others},
year = {2023},
eprint = {2310.08864},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2310.08864},
}
Opens a 50,000-hour annotated egocentric video corpus with pipelines and checkpoints. Reports a log-linear law between pretraining hours and robot success, and 84.6% on RoboTwin 2.0 Clean2Random.开源 5 万小时带标注的第一视角视频语料及流程和权重。报告预训练时长与机器人成功率之间的对数线性规律，后训练后在 RoboTwin 2.0 Clean2Random 上达到 84.6%。
@misc{ma2026open,
title = {{Open-HumanWAM: An Open Recipe for Scaling Human and Robot Experience for Embodied Foundation Models}},
author = {Juncheng Ma and Jiaxin Li and Jiankai Tu and Bo Liang and Guodong Lv and Yufan Deng and Junzhe Tang and Liangzhu Yang and Chen Bian and Zicong Cao and Baohao Zhu and Jianxin Bi and Tengfei Yan and Yinyuan Zhang and Kewei Zhang and Xinglong Chen and Kaiqi Chen and Duomin Wang and Xiaotian Tang and Liang Tan and Li Li and Yuan Xu and Haosen Yang and Yuqi Wang and Jingfei Xia and Songyan Huang and Zhiyang Dou and Zhen Dong and Enze Xie and Tat-Seng Chua and Daquan Zhou},
year = {2026},
eprint = {2611.06817},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2611.06817},
}
Learned from 16,151 hours of handheld gripper demos with no robot data; reports 78% average success over ten tasks versus 27% for the best other VLA, and releases the dataset, weights and post-training recipe.从 16,151 小时手持夹爪演示中学习，全程不采机器人数据；报告十项任务平均成功率 78%，其他 VLA 最好为 27%，并开放数据集、权重和后训练方案。
@misc{wu2026airoa,
title = {{AIRoA-0: Long-Horizon, Precise, Cross-Embodiment Manipulation from Handheld Data}},
author = {Yueh-Hua Wu and Kei Ota and Hanlong Li and Hayata Yuasa and Jason Wang and Kengo Ikeuchi and Kensuke Chiyoma and Makoto Sugiura and Masafumi Takahashi and Masatoshi Tateno and Naoaki Kanazawa and Nicola Scianca and Petr Khrapchenkov and Satoshi Otsubo and Takehiko Ohkawa and Takuya Okubo and Tatsuya Matsushima and Yuki Noguchi and Yuma Shin and Zixuan Zhong},
year = {2026},
eprint = {2611.03815},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.03815},
}
@misc{chi2024universal,
title = {{Universal Manipulation Interface: In-The-Wild Robot Teaching Without In-The-Wild Robots}},
author = {Cheng Chi and Zhenjia Xu and Chuer Pan and Eric Cousineau and Benjamin Burchfiel and Siyuan Feng and Russ Tedrake and Shuran Song},
year = {2024},
eprint = {2402.10329},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2402.10329},
}
PyTorch library with datasets, pretrained policies (ACT, Diffusion, SmolVLA) and drivers for low-cost arms.PyTorch 机器人学习库，提供数据集、预训练策略（ACT、Diffusion、SmolVLA）与低成本机械臂驱动。
@misc{liu2023libero,
title = {{LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning}},
author = {Bo Liu and Yifeng Zhu and Chongkai Gao and Yihao Feng and Qiang Liu and Yuke Zhu and Peter Stone},
year = {2023},
eprint = {2306.03310},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2306.03310},
}
CoRL 2026 Spotlight. Simulation benchmark for wires and cables in industrial assembly, aimed at the gap left by benchmarks tied to specific hardware or lacking modular, customizable task assets.CoRL 2026 Spotlight。面向工业装配中线缆等可变形线状物体（DLO）的仿真基准，针对现有基准绑定特定硬件、缺少模块化可定制任务资产等问题。
@misc{zhu2026wirecraft,
title = {{WireCraft: A Simulation Benchmark for Industrial DLO Manipulation}},
author = {Chongyu Zhu and Ramy ElMallah and Hyegang Kim and Zachary Tang and Jiachen Rao and Artem Arutyunov and Seungyeon Ha and Chi-Guhn Lee},
year = {2026},
eprint = {2606.18097},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.18097},
}
1,000 expert-curated questions over 11 image and video task types test multimodal LLMs as the brain of a self-driving chemistry lab: affordances, trajectories, apparatus setup, safety and step reasoning.1000 道专家整理的题目、11 种图像和视频任务，检验多模态大模型能否担任自动化化学实验室的“大脑”：可供性、轨迹、仪器搭建、安全与步骤推理。
@misc{lin2026beaker,
title = {{BEAKER: An Expert-Curated Benchmark for Embodied Brains in Self-Driving Chemical Laboratories}},
author = {Fei Lin and Tengchao Zhang and Ziyang Gong and Xiaotong Yu and Bohan Zhang and Yifan Zhou and Qihao Yang and Ji Dai and Dong Li and Yining Jiang and Qinghua Ni and Jun Huang and Qiang Zhang and Yue Zhou and Zhihang Zhong and Yonglin Tian and Fei-Yue Wang and Xue Yang},
year = {2026},
eprint = {2611.06911},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2611.06911},
}
Simulation pipeline that writes and checks task code with multimodal LLMs to mass-produce bimanual demos under heavy randomization (clutter, lighting, backgrounds, instructions); also a 50-task dual-arm benchmark.用多模态大模型自动生成并校验任务代码，在强域随机化（杂物、光照、背景、指令）下批量产出双臂演示数据的仿真管线，同时提供 50 个双臂任务的基准。
@misc{chen2025robotwin,
title = {{RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation}},
author = {Tianxing Chen and Zanxin Chen and Baijun Chen and Zijian Cai and Yibin Liu and Zixuan Li and Qiwei Liang and Xianliang Lin and Yiheng Ge and Zhenyu Gu and Weiliang Deng and Yubin Guo and Tian Nian and Xuanbing Xie and Qiangyu Chen and Kailun Su and Tianling Xu and Guodong Liu and Mengkang Hu and Huan-ang Gao and Kaixuan Wang and Zhixuan Liang and Yusen Qin and Xiaokang Yang and Ping Luo and Yao Mu},
year = {2025},
eprint = {2506.18088},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2506.18088},
}
MetaSim puts many simulators (Isaac Sim/Gym, MuJoCo, Genesis, SAPIEN…) behind one interface; adds a large synthetic dataset migrated from public benchmarks and unified IL/RL benchmarks.MetaSim 用统一接口封装多种仿真器（Isaac Sim/Gym、MuJoCo、Genesis、SAPIEN、PyBullet 等）；提供由公开基准迁移而来的大规模合成数据集，以及按泛化层级划分的统一模仿学习/强化学习基准。
@misc{geng2025roboverse,
title = {{RoboVerse: Towards a Unified Platform, Dataset and Benchmark for Scalable and Generalizable Robot Learning}},
author = {Haoran Geng and Feishi Wang and Songlin Wei and Yuyang Li and Bangjun Wang and Boshi An and Charlie Tianyue Cheng and Haozhe Lou and Peihao Li and Yen-Jen Wang and Yutong Liang and Dylan Goetting and Chaoyi Xu and Haozhe Chen and Yuxi Qian and Yiran Geng and Jiageng Mao and Weikang Wan and Mingtong Zhang and Jiangran Lyu and Siheng Zhao and Jiazhao Zhang and Jialiang Zhang and Chengyang Zhao and Haoran Lu and Yufei Ding and Ran Gong and Yuran Wang and Yuxuan Kuang and Ruihai Wu and Baoxiong Jia and Carlo Sferrazza and Hao Dong and Siyuan Huang and Yue Wang and Jitendra Malik and Pieter Abbeel},
year = {2025},
eprint = {2504.18904},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2504.18904},
}
120 kitchen scenes, 2,500+ objects and 100 tasks with generated demonstrations for training generalist policies.120 个厨房场景、2500 多个物体与 100 个任务，并附生成演示，用于训练通用策略。
@misc{nasiriany2024robocasa,
title = {{RoboCasa: Large-Scale Simulation of Everyday Tasks for Generalist Robots}},
author = {Soroush Nasiriany and Abhiram Maddukuri and Lance Zhang and Adeet Parikh and Aaron Lo and Abhishek Joshi and Ajay Mandlekar and Yuke Zhu},
year = {2024},
eprint = {2406.02523},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2406.02523},
}
Simulated interactive 3D society for many robot types; its GRScenes set has 100k annotated scenes across 89 categories that combine into city-scale worlds.面向多种机器人的交互式三维仿真社会；GRScenes 含 10 万标注场景、89 个类别，可组合成城市级环境。
@misc{wang2024grutopia,
title = {{GRUtopia: Dream General Robots in a City at Scale}},
author = {Hanqing Wang and Jiahe Chen and Wensi Huang and Qingwei Ben and Tai Wang and Boyu Mi and Tao Huang and Siheng Zhao and Yilun Chen and Sizhe Yang and Peizhou Cao and Wenye Yu and Zichao Ye and Jialun Li and Junfeng Long and Zirui Wang and Huiling Wang and Ying Zhao and Zhongying Tu and Yu Qiao and Dahua Lin and Jiangmiao Pang},
year = {2024},
eprint = {2407.10943},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2407.10943},
}
@misc{li2024evaluating,
title = {{Evaluating Real-World Robot Manipulation Policies in Simulation}},
author = {Xuanlin Li and Kyle Hsu and Jiayuan Gu and Karl Pertsch and Oier Mees and Homer Rich Walke and Chuyuan Fu and Ishikaa Lunawat and Isabel Sieh and Sean Kirmani and Sergey Levine and Jiajun Wu and Chelsea Finn and Hao Su and Quan Vuong and Ted Xiao},
year = {2024},
eprint = {2405.05941},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2405.05941},
}
1,000 household activities chosen from a survey of what people want robots to do, defined in a logic language and simulated in OmniGibson with realistic physics and rendering.依据人们希望机器人做什么的调查挑选出 1000 项家务活动，用逻辑语言定义，并在具备真实物理与渲染的 OmniGibson 中仿真。
@misc{li2024behavior,
title = {{BEHAVIOR-1K: A Human-Centered, Embodied AI Benchmark with 1,000 Everyday Activities and Realistic Simulation}},
author = {Chengshu Li and Ruohan Zhang and Josiah Wong and Cem Gokmen and Sanjana Srivastava and Roberto Martín-Martín and Chen Wang and Gabrael Levine and Wensi Ai and Benjamin Martinez and Hang Yin and Michael Lingelbach and Minjune Hwang and Ayano Hiranaka and Sujay Garlanka and Arman Aydin and Sharon Lee and Jiankai Sun and Mona Anvari and Manasi Sharma and Dhruva Bansal and Samuel Hunter and Kyu-Young Kim and Alan Lou and Caleb R Matthews and Ivan Villa-Renteria and Jerry Huayang Tang and Claire Tang and Fei Xia and Yunzhu Li and Silvio Savarese and Hyowon Gweon and C. Karen Liu and Jiajun Wu and Li Fei-Fei},
year = {2024},
eprint = {2403.09227},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2403.09227},
}
robomimic — a controlled study of imitation and offline RL on human-collected manipulation demos (demo quality, observation choices, history-aware BC), released with the framework and datasets.robomimic：在人类采集的操作示教数据上系统比较模仿学习与离线强化学习（示教质量、观测选择、带历史的行为克隆），并开源框架与数据集。
@misc{mandlekar2021what,
title = {{What Matters in Learning from Offline Human Demonstrations for Robot Manipulation}},
author = {Ajay Mandlekar and Danfei Xu and Josiah Wong and Soroush Nasiriany and Chen Wang and Rohun Kulkarni and Li Fei-Fei and Silvio Savarese and Yuke Zhu and Roberto Martín-Martín},
year = {2021},
eprint = {2108.03298},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2108.03298},
}
@misc{mees2021calvin,
title = {{CALVIN: A Benchmark for Language-Conditioned Policy Learning for Long-Horizon Robot Manipulation Tasks}},
author = {Oier Mees and Lukas Hermann and Erick Rosete-Beas and Wolfram Burgard},
year = {2021},
eprint = {2112.03227},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2112.03227},
}
@misc{zhu2020robosuite,
title = {{robosuite: A Modular Simulation Framework and Benchmark for Robot Learning}},
author = {Yuke Zhu and Josiah Wong and Ajay Mandlekar and Roberto Martín-Martín and Abhishek Joshi and Kevin Lin and Abhiram Maddukuri and Soroush Nasiriany and Yifeng Zhu},
year = {2020},
eprint = {2009.12293},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2009.12293},
}
@misc{james2019rlbench,
title = {{RLBench: The Robot Learning Benchmark \& Learning Environment}},
author = {Stephen James and Zicong Ma and David Rovick Arrojo and Andrew J. Davison},
year = {2019},
eprint = {1909.12271},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/1909.12271},
}
@misc{yu2019meta,
title = {{Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learning}},
author = {Tianhe Yu and Deirdre Quillen and Zhanpeng He and Ryan Julian and Avnish Narayan and Hayden Shively and Adithya Bellathur and Karol Hausman and Chelsea Finn and Sergey Levine},
year = {2019},
eprint = {1910.10897},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/1910.10897},
}
@misc{savva2019habitat,
title = {{Habitat: A Platform for Embodied AI Research}},
author = {Manolis Savva and Abhishek Kadian and Oleksandr Maksymets and Yili Zhao and Erik Wijmans and Bhavana Jain and Julian Straub and Jia Liu and Vladlen Koltun and Jitendra Malik and Devi Parikh and Dhruv Batra},
year = {2019},
eprint = {1904.01201},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/1904.01201},
}
RGB-D scans of 1,500+ indoor scenes with 3D reconstructions, camera poses and semantic labels; the standard benchmark for 3D scene understanding that embodied perception builds on.1500+ 个室内场景的 RGB-D 扫描，含三维重建、相机位姿与语义标注；具身感知所依赖的三维场景理解标准基准。
@misc{dai2017scannet,
title = {{ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenes}},
author = {Angela Dai and Angel X. Chang and Manolis Savva and Maciej Halber and Thomas Funkhouser and Matthias Nießner},
year = {2017},
eprint = {1702.04405},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/1702.04405},
}
S² is a training-free offline metric of how well a model's representation aligns with physical motion. Reports it tracks robot task success across model families and can rank VLM backbones before VLA fine-tuning.S² 是一种无需训练的离线指标，衡量模型视觉表征与物理运动几何的对齐程度（把柏拉图表征假说扩展到机器人）；报告它与多类模型的真实任务成功率正相关，可在微调成 VLA 之前筛选 VLM 骨干。
@misc{xie2026sensorimotor,
title = {{Sensorimotor Alignment: A Platonic Proxy for Embodied Policy Success}},
author = {Yunfei Xie and Howard H. Qian and Shiyi Lan and Kaiyu Hang and Chen Wei},
year = {2026},
eprint = {2610.31670},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2610.31670},
}
Paired test of previously proposed inference and data-collection tricks on all 100 BEHAVIOR-1K tasks with a π0.5 based on the 2025 challenge winner (Q-score 0.39): the tricks do not transfer uniformly.在 BEHAVIOR-1K 全部 100 个任务上，以基于 2025 BEHAVIOR 挑战赛冠军方案微调的 π0.5（公开测试集 Q 分 0.39）为对象，对已有推理与数据采集技巧做配对评测；结论是这些技巧在新环境与任务上的效果并不一致，提醒小规模结论不宜直接外推。
@misc{hammele2026what,
title = {{What Actually Matters? A Paired Evaluation of Inference Settings for π0.5 on BEHAVIOR-1K}},
author = {Ruben Hammele and Samira Huber and Mohammad Nazeri and Amirreza Payandeh and Alexandyr T. Card and Aaron Nguyen and Anuj Pokhrel and Daeun Song and Xuesu Xiao and Sören Pirk},
year = {2026},
eprint = {2610.37917},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.37917},
}
Rates VLA and world-action models on five axes (capability, training and deployment cost, verified research value, reproducibility), not success rate alone; withholds overall scores when evidence is missing.多维度评测 VLA 与世界-动作模型（WAM）：除任务能力外，还衡量下游训练成本、部署成本、经本地验证的研究价值和复现证据，按不同算力预算给出 0–100 画像；证据缺失时明确标注、不给总分，并开源评分代码与机器可读证据。
@misc{ma2026roboseed,
title = {{RoboSeed: Multidimensional Evaluation of Vision-Language-Action and World-Action Models: Capability, Resource Affordability, and Research Evidence}},
author = {Xin Ma and Furong Chen and Xinlong Man and Bo Zhang and Yalong Guo and Haolan Lei and Chaochen Cui and Hongyuan Zhang and Mulin Chen and Yi Wang and Lap-pui Chau and Chuang Yang},
year = {2026},
eprint = {2610.40026},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.40026},
}
Memory for VLAs that keeps only sparse keyframes the policy itself predicts will matter later. Adds the RoboTwin-MeM benchmark; reports +40% average success over memory-augmented VLAs across 17 sim and 4 real tasks.为 VLA 加入稀疏记忆：由策略自己预测哪些关键帧之后会有用并只保存这些帧。同时提出 RoboTwin-MeM 基准；报告在 17 个仿真和 4 个真机任务上平均成功率比带记忆的 VLA 高 40%。
@misc{yang2026eventvla,
title = {{EventVLA: Event-Driven Visual Evidence Memory for Long-Horizon Vision-Language-Action Policies}},
author = {Ganlin Yang and Zhangzheng Tu and Yuqiang Yang and Sitong Mao and Junyi Dong and Tianxing Chen and Jiaqi Peng and Jing Xiong and Jiafei Cao and Jifeng Dai and Wengang Zhou and Yao Mu and Tai Wang},
year = {2026},
eprint = {2606.20092},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2606.20092},
}
Converts human videos from five sources into robot-aligned observations and actions (about 630K episodes). Pretraining a VLA on more of it raises real-task completion from 51.5% to 80.3% and OOD from 34.9% to 72.2%.把五个来源的人类视频转换成与机器人对齐的观测和动作（约 63 万条轨迹）。报告用更多此类数据预训练 VLA，真机任务完成率从 51.5% 升到 80.3%，分布外从 34.9% 升到 72.2%。
@misc{jeong2026huro,
title = {{HuRo: Robotizing Human Videos for Scalable VLA Pretraining}},
author = {Jinho Jeong and Se June Joo and Jaehyun Kang and Dongyun Kim and Yena Kim and Hanjung Kim and Seon Joo Kim},
year = {2026},
eprint = {2609.10706},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.10706},
}
Joint VLA pretraining on 4.53K h of robot/sim data plus 1.48K h of egocentric human video with pseudo-actions, camera-space actions and a reliability-weighted loss. Reports SOTA on RoboCasa GR1 TableTop and RoboTwin 2.0.用 4530 小时机器人/仿真数据加 1480 小时带伪动作的第一视角人类视频联合预训练 VLA，采用相机空间动作与按可靠性加权的损失。报告在 RoboCasa GR1 TableTop 和 RoboTwin 2.0 上达到最优。
@misc{li2026ace,
title = {{ACE-Ego-0: Unifying Egocentric Human and Robotic Data for VLA Pretraining}},
author = {Hao Li and Ganlong Zhao and Yufei Liu and Haotian Hou and Guoquan Ye and Tongyan Fang and Chunxiao Liu and Siyuan Huang and Jianbo Liu and Xiaogang Wang and Hongsheng Li},
year = {2026},
eprint = {2606.17200},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.17200},
}
Lightweight probe on a VLA's latent layers that flags failures early with calibrated miss and false-alarm bounds and names the failure mode. Releases 3,180 annotated manipulation episodes.在 VLA 隐层上加轻量探针，提前预警失败并给出校准的漏报与误报上界，还能判断失败类型。并发布 3180 条带标注的操作轨迹。
@misc{podolinsky2026failure,
title = {{Failure Prediction and Classification in Vision-Language-Action Models Via Calibrated Latent Probing}},
author = {Michal P. Podolinsky and Yunhao Yang and Daniel Little and Hong Lu and Rohan Siva and Olin Goudey and Neel P. Bhatt and Matthias Scheutz and Ufuk Topcu},
year = {2026},
eprint = {2611.08425},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.08425},
}
One two-hand policy (PPO + BC on its own verified successes) tracks retargeted human hand-object motions; ships BiHOI (1,575 sequences, 229 objects). Reports 86.8% on unseen trajectories, 57.3% on new objects.用一个双手跟踪策略（PPO 结合对自身成功执行的行为克隆）复现重定向后的人手-物体交互动作，并发布 BiHOI 数据集（1575 段双手序列、229 个物体）。论文报告在未见轨迹上成功率 86.8%、新物体上 57.3%。
@misc{sun2026donic,
title = {{DONIC: Generalist Dexterous Bimanual Tracking Control from Human Demonstrations}},
author = {Mingze Sun and Ziang Cao and Yuan Fei and Qiankuan Tao and Zihan Wang and Tianqi Liu and Junhao Chen and Haiwen Diao and Fangzhou Hong and Zhaoxi Chen and Ziwei Liu},
year = {2026},
eprint = {2610.29997},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.29997},
}
36 contact-rich tasks built around eight everyday skills, with 3D-printable objects that each have a matching sim model. Reports that tested visuo-tactile policies did not clearly beat vision-only ones.围绕八种日常接触技能设计的 36 个接触密集型任务，40 多个物体可用消费级 3D 打印机复现，且每个都有对应仿真模型。报告称所测的视触觉策略并未明显优于纯视觉策略。
MimicGen-style demo synthesis that respects multi-arm constraints: plans and schedules skills across arms. Reports 79.6% generation success on 28 sim variants and 50-80% real success via zero-shot sim-to-real.面向双臂/多臂机器人的示范自动生成：在少量源示范基础上联合做规划与技能调度，显式处理多臂间约束，避免现有生成方法产生的碰撞与失败。28 个仿真任务变体上数据生成成功率 79.6%，训练出的策略经零样本 sim-to-real 在 8 个真实变体上成功率 50–80%。
@misc{jawale2026schedulegen,
title = {{ScheduleGen: Demonstration Generation via Skill Scheduling for Multi-Arm Imitation Learning}},
author = {Neel Jawale and Stephan Pleines and Soha Pouya and Yuke Zhu and Ajay Mandlekar and Caelan Garrett},
year = {2026},
eprint = {2610.39429},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.39429},
}
5B world action model pretrained on 605 h of RGB-D UMI data that predicts RGB, metric depth and actions. Adds RoboTwin-UMI; reports 85% average real success, +24.2 points over the best baseline.在 605 小时 RGB-D UMI 数据上预训练的 5B 世界动作模型，同时预测 RGB、度量深度和动作。并提出 RoboTwin-UMI 基准；报告真机平均成功率 85%，比最强基线高 24.2 个百分点。
@misc{yuan2026spatial,
title = {{Spatial-UMI-Dream: Scaling World Action Models with Metric Geometry for UMI Manipulation}},
author = {Weiduo Yuan and Ze Yuan and Sihan Liu and Jionghao Wang and Liang Heng and Zhenyu Zhao and Yue Wang},
year = {2026},
eprint = {2611.07229},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.07229},
}
Drops an obstacle into a fixed grasp's approach path and lets a planner relabel it: eight feasibility predictors fall from 0.92–0.996 AUROC to chance, so they ignore collision geometry. Labels and pipeline released.固定抓取位姿、在接近路径中插入障碍物并由规划器重新标注：八种可行性预测器的 AUROC 从 0.92–0.996 跌至随机水平，说明它们并未利用碰撞几何。发布标签与流程。
128 simulated tactile tasks in three regimes (Precise, Recognition, Dynamic) with 12 perturbations, plus a tactile world model that treats touch as an evolving state and refines actions with predicted contact.128 个仿真触觉任务，分精确、识别、动态三类，含 12 种扰动；并提出动作条件的触觉世界模型，把触觉当作随时间演化的状态，用预测的触觉信息修正动作。
@misc{shi2026tacmind,
title = {{TacMind: Benchmarking and Modeling Tactile Interaction States for Robotic Manipulation}},
author = {Jiaxin Shi and Zibin Wu and Yichi Zhang and Jianghua Duan and Siyu Zhu and Weihao Yuan},
year = {2026},
eprint = {2611.00455},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.00455},
}
HumanVerse, 500 h of egocentric human video with body and hand motion, aligned to the Unitree G1 for mid-training a humanoid model; reports 90% on SIMPLE and 62.5% on real loco-manipulation tasks.构建 500 小时的第一视角人类数据集 HumanVerse（含同步的身体与手部动作），对齐到 Unitree G1 后用于人形基础模型 λ0 的中期训练；论文报告 SIMPLE 上成功率 90%、真实移动操作任务 62.5%，人类数据越多下游表现越好。
@misc{team2026scaling,
title = {{$λ_0$: Scaling Open-World Egocentric Human Data for Humanoid Loco-Manipulation}},
author = {Lambda-humanoid Team and : and Chongyang Xu and Jin Chen and Jinhui Ye and Yilun Chen and Yiming Jiang and Zhao Wu and Hu Wei and Lin Qu and Shifeng Zhang and Yisheng He and Yuming Jiang and Zenan Meng and Zheng Wu and Mu Xu and Steven Hoi},
year = {2026},
eprint = {2610.18950},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.18950},
}
Humanoid that reacts to a human partner: a token model generates G1 whole-body motion conditioned on the partner, and a tracker executes it. Adds a 260-hour human-humanoid interaction benchmark.让人形机器人对人类伙伴做出回应：token 化序列模型以伙伴动作为条件生成 G1 全身动作，再由跟踪控制器执行。并提供 260 小时的人-人形交互基准。
@misc{maluleke2026social,
title = {{Social Humanoid: Socially Responsive Human-Humanoid Interaction}},
author = {Vongani H. Maluleke and James Ni and Kie Horiuchi and Lea Wilken and Jiahui Lei and Evonne Ng and Roei Herzig and Jitendra Malik and Trevor Darrell and Angjoo Kanazawa},
year = {2026},
eprint = {2611.08407},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.08407},
}
Training-free physics check for video models: an MLLM states history-based expectations and PhysTrace marks when the video breaks them; adds IntPhys2-Timing (reports up to 70.7% hit rate).无需训练的视频模型物理一致性评测：由多模态大模型根据历史帧给出物理预期，再定位视频中违背预期的时间段并给出可解释描述；新增带违规区间标注的 IntPhys2-Timing（报告命中率最高 70.7%），面向依赖视频模型的世界动作模型。
@misc{yoo2026phystrace,
title = {{PhysTrace: Interpretable Temporal Evaluation of Physical Consistency in Video Generative Models}},
author = {Taekoan Yoo and Wonkyung Jung and Kyunghun Kim and Kyeongbo Kong},
year = {2026},
eprint = {2610.31429},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2610.31429},
}
Generative world model of 3D particle and rigid-body motion covering rigid, deformable, cloth and granular materials. Reports 31% lower error than prior learned dynamics; planning searches its latent noise.生成式 3D 粒子与刚体运动世界模型，统一覆盖刚体、可变形体、布料与颗粒材料。报告预测误差比此前最好的学习动力学低 31%，并通过在潜噪声空间搜索来做规划。
@misc{ren2026gpwm,
title = {{GPWM: Modeling the World as 3D Particle Motion}},
author = {Robert Ren and Matthew Bronars and He Zhu and Yikai Tang and Jianing Zhang and Sergey Zakharov and Pavel Tokmakov and Katerina Fragkiadaki},
year = {2026},
eprint = {2611.08338},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.08338},
}
Recalibrates DROID into DROID-3D with metric depth and multi-view extrinsics, then trains a video world model that predicts RGB and depth. Depth supervision reportedly also improves RGB by 1.48 dB PSNR.把 DROID 重新标定为带度量深度与多视角外参的 DROID-3D，并训练同时预测 RGB 与深度的视频世界模型。报告深度监督还让 RGB 预测提升 1.48 dB PSNR。
@misc{bardhan2026depthworld,
title = {{DepthWorld: 3D World Model for Robot Manipulation}},
author = {Jai Bardhan and Josef Sivic and Vladimir Petrik},
year = {2026},
eprint = {2610.08780},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.08780},
}
Factorizes an interaction into action, camera and embodiment so one human video can be re-rendered as many robot rollouts. Generated data helps on RoboCasa GR1 and a real IRON humanoid.把一次交互分解为动作、相机与本体三个因素，使一段人类视频可重新生成为多种机器人视角的数据。生成数据在 RoboCasa GR1 与真实 IRON 人形上带来提升。
@misc{chen2026anyworld,
title = {{AnyWorld: Factorized Egocentric World Models for Cross-Embodiment Generalization}},
author = {Cheng Chen and Jerry Bai and Jiacheng Wei and Boyu Chen and Xiaoji Zheng and Fan Wu and Minghao Yang and Tianrun Chen and Ruibo Li and Xiaoyu Yue and Xiaoyang Guo and Yixiao Ge and Guosheng Lin and Fayao Liu},
year = {2026},
eprint = {2608.29242},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2608.29242},
}
Represents object motion as dense 3D trajectories over the whole surface, occluded parts included, for rigid and deformable objects. Adds a 1,210-hour flow dataset and a world model predicting it from image + task.把物体运动表示为覆盖整个表面（含遮挡部分）的稠密 3D 轨迹，适用于刚体与可形变物体。发布 1210 小时的全物体流数据集，并训练从图像和任务描述预测该流的世界模型。
@misc{chen2026objectworld,
title = {{ObjectWorld: Scaling Object-Centric World Models with A Large-Scale 3D Full-Object Flow Dataset}},
author = {Haonan Chen and Sicheng Isabella Lu and Haitian Li and Jiajin Wu and Hexiang Wei and Jiawei Zhang and Xin Fei and Yiwen Hou and Chenrui Tie and Chongkai Gao and Sheng Wang and Lin Shao},
year = {2026},
eprint = {2611.11085},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.11085},
}
Simulator with embodied NPCs, dialogue and event-driven task states, plus a skill harness for frozen MLLMs; its 360-episode bench reports general MLLMs can beat embodied ones; task updates and HRI stay hard.带具身 NPC、受限对话和事件驱动任务状态机的物理仿真框架，并提供让冻结多模态大模型调用多级技能的执行框架；基于 30 个室内场景构建 360 回合基准，评测 8 个模型，报告通用模型可胜过具身专用模型，任务变更和人机协作仍是难点。
@misc{xu2026tarsphere,
title = {{TARSphere: A Robot-Oriented Full-Stack Framework for Long-Horizon Embodied Agents in Evolving Physical and Social Environments}},
author = {Yicheng Xu and Zhikai Xu and Ziyuan Chen and Yuhang Dong and Tao Zhang and Zhucun Xue and Yong Liu and Jiangning Zhang},
year = {2026},
eprint = {2610.30043},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.30043},
}
200K+ navigation-planning samples annotated from camera trajectories up to instructions; RL post-training inside a video world model, used as a cheap stand-in environment, reportedly helps on unseen scenes.包含 20 万余条导航规划样本的数据集，场景多样，标注从相机运动轨迹到高层语义指令；并把视频世界模型当作低成本交互环境做强化学习后训练，报告称在未见场景上有明显提升。
@misc{feng2026embodiedworld,
title = {{EmbodiedWorld-200K: Toward Embodied World Agents via Embodied-Planning Dataset and Interactive World Models}},
author = {X. Feng and J. Tang and Z. Lin and L. Chen and S. Shao and Q. Lu and K. Huang},
year = {2026},
eprint = {2610.25959},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2610.25959},
}
GEMS keeps a 3D Gaussian Splatting map that is updated only where the scene changes, and an agent renders views on demand to answer when/where questions. Introduces GEMS-Bench (15 scenes, 473 QA pairs).GEMS 维护一张只在场景变化处增量更新的 3DGS 地图，智能体按需渲染新视角来回答时空问题和定位物体；同时发布 GEMS-Bench（3 个真实 + 12 个合成环境，473 个问答对），报告在物体定位和时空问答上优于近期基线。
@misc{chen2026gaussian,
title = {{Gaussian Episodic Memory for Long-Term Spatiotemporal Reasoning}},
author = {Ling-Chia Chen and Abdelrhman Werby and Kaan Aytekin and Kai O. Arras},
year = {2026},
eprint = {2610.33158},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2610.33158},
}
OSMA-Bench: matched scenes that differ only in obstacle type test whether navigation policies slow down and keep clearance by how risky an obstacle is. VLM and VLA policies recognise the risk but rarely act on it.OSMA-Bench 用仅障碍物类型不同的配对场景，检验导航策略能否按障碍物危险程度调整间距与速度，并提出语义敏感度指数 SSI。结果显示 VLM 与 VLA 策略能识别哪些障碍更危险，却很少把这种判断体现在实际运动中；在 VLM 方案中加入显式参数推断步骤有所改善。
@misc{bae2026benchmarking,
title = {{Benchmarking Obstacle-Conditioned Safety in Household Robot Navigation}},
author = {Hyunwoong Bae and Jisu Han and Jiyeon Choi and Dohoon Kim and Taesup Moon},
year = {2026},
eprint = {2610.34674},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.34674},
}
Adds simulated odour transport to Habitat-Sim using Fire Dynamics Simulator, giving a smell-and-vision navigation benchmark for finding invisible gas leaks and early-stage fires, with a multimodal RL baseline.用 Fire Dynamics Simulator 为 Habitat-Sim 加入气味传输仿真，构成视觉加嗅觉的导航基准，任务是定位不可见的气体泄漏和早期火源，并提供多模态强化学习基线。
@misc{lee2026odorspaces,
title = {{OdorSpaces: Visual-Olfactory Embodied Navigation in 3D Environments}},
author = {Changyeon Lee and Kelin Yu and Zheng Wei and Shuna Ni and Ruohan Gao},
year = {2026},
eprint = {2611.05738},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.05738},
}
Scores delivery-robot navigation by money: cost, revenue and break-even, using Isaac Sim plus public filings and injury data. Reports that none of seven baselines is economically viable yet.以经济账评估配送机器人导航：在 Isaac Sim 中结合公开财报与伤害数据计算成本、收入与盈亏平衡。报告七种基线方法目前都无法盈利。
@misc{seong2025costnav,
title = {{CostNav: A Navigation Benchmark for Real-World Economic-Cost Evaluation of Physical AI Agents}},
author = {Haebin Seong and Sungmin Kim and Yongjun Cho and Myunchul Joe and Geunwoo Kim and Yubeen Park and Sunhoo Kim and Samwoo Seong and Yoonshik Kim and Suhwan Choi and Jaeyoon Jung and Jiyong Youn and Jinmyung Kwak and Sunghee Ahn and Jaemin Lee and Younggil Do and Seungyeop Yi and Woojin Cheong and Minhyeok Oh and Minchan Kim and Seongjae Kang and Youngjae Yu and Yunsung Lee},
year = {2025},
eprint = {2511.20216},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2511.20216},
}
PhysRSI-eval scores VLM agents on 18 manipulation tasks under a fixed trial budget on five axes, incl. learning from failure; reports similar benchmark averages hide different improvement speeds.PhysRSI-eval 在固定交互预算下用 18 个操作任务、五个维度（含从失败中改进）评测 VLM 智能体；报告公开基准平均分相近的模型改进速度差别很大，且早期物理反馈可用于预测后续进展。
@misc{li2026where,
title = {{Where is Physical RSI capability from? An Observational Scaling Law of LLM-as-policy}},
author = {Zekai Li and Jiayi Yu and Jinlu Zhang and Luyang Niu and Zekun Qi and Jun Zheng and Xuchuan Chen and Dairu Liu and Yintianrun Zhang and He Wang and Li Yi},
year = {2026},
eprint = {2610.26848},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.26848},
}
Sanity check for demo-curation methods on mixed-skill robomimic data: one score ranks demos almost exactly like episode length, and after Holm correction no curated subset significantly beats training on all data.对示范数据筛选方法做严格对照：在混合技能的 robomimic 数据上与等量随机子集和全量训练比较，发现某评分几乎等同于按轨迹长度排序；经多重比较校正后，没有任何筛选条件显著优于全量数据训练（CoRL 2026 Workshop 论文）。
@misc{xu2026curaeval,
title = {{CuraEval: Evaluating Demonstration Curation on Mixed-Skill Robot Data}},
author = {Jiuyi Xu and Meida Chen and Yang Ye and Yangming Shi},
year = {2026},
eprint = {2610.38221},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.38221},
}
Compresses policy rollouts into one directed graph of behavior modes linked to outcomes, to guide what data to collect next. In a user study on three real tasks, 16 of 18 participants preferred it to rollout videos.把策略回放压缩成一张有向图，将行为模式与结果相连，帮助决定下一步采集什么数据。在三个真实任务的用户研究中，18 名参与者有 16 名更喜欢它而非回放视频。
@misc{bauer2026behavior,
title = {{Behavior Graphs: Unsupervised Graph Structures for Interpreting Robot Deployment Data}},
author = {Erik Bauer and Jakob Thumm and Christopher Agia and Rohan Sinha and Chen Xu and Haruki Nishimura and Jeannette Bohg and Marco Pavone and Masha Itkina},
year = {2026},
eprint = {2611.00405},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.00405},
}
Training-free progress reward: reads a video-language model's token probability that the instructed task is done, rather than asking it for a number. Introduces ManiRewardBench (130 tasks, four robots).免训练的进度奖励：不让视频语言模型输出数值，而是读取其判断“任务已完成”的 token 概率作为稠密奖励。并提出 ManiRewardBench（130 个任务、4 种机器人）。
@misc{chen2026topreward,
title = {{TOPReward: Token Probabilities as Hidden Zero-Shot Rewards for Robotics}},
author = {Shirui Chen and Cole Harrison and Ying-Chun Lee and Angela Jin Yang and Zhongzheng Ren and Lillian J. Ratliff and Jiafei Duan and Dieter Fox and Ranjay Krishna},
year = {2026},
eprint = {2602.19313},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2602.19313},
}
Scores how much a small action error at each demo state would hurt the outcome, using an adversarial disturbance policy. The scores reweight behavior cloning and select smaller, higher-value demo subsets.用对抗扰动策略衡量每个示教状态上的小动作误差会对结果造成多大损害。该分数可用于给行为克隆加权，并挑选更小但价值更高的示教子集。
@misc{ciftci2026where,
title = {{Where Small Mistakes Matter: Quantifying State Criticality for Imitation Learning}},
author = {Yusuf Umut Ciftci and Joonwon Kang and Ajay Suresha Sathya and Somil Bansal},
year = {2026},
eprint = {2610.17709},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2610.17709},
}
@misc{houssaini2026dexsuite,
title = {{DexSuite: An All-in-One Framework for Dexterous Manipulation Across Arms, Hands, and Tasks}},
author = {Anas Houssaini and Hugo He and Junming Shi and Amin Abyaneh and Ricardo Chahine and Shuo Wen and Mohamad H. Danesh and Charles-Étienne Fréchette and Mariana Sosa Guzman and Mathias Desrochers and Rayan Houssaini and Adam Landa and Theophile Soulie and Charlotte Morissette and Annie Tims and Martino Russi and Jonathan Lussier and Gregory Dudek and Doina Precup and Julen Urain and Hsiu-Chin Lin and David Meger},
year = {2026},
eprint = {2610.37799},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.37799},
}
Coding agents build robotics simulations, checked by an adversarial create-and-audit loop; result RoboWorld: 25 sims chosen by U.S. GDP share, 200 RL envs. Best frontier models average just 0.06 reward on test tasks.把“写仿真”当作编码智能体任务：智能体生成机器人仿真与任务，再由对抗式的创建-审核循环把关；据此构建 RoboWorld（按美国 GDP 构成挑选的 25 个仿真、200 个强化学习环境），最强前沿模型在测试任务上平均奖励仅 0.06。
@misc{aggarwal2026gym,
title = {{Gym-Anything for Robotics: Scaling physical simulations with digital agents}},
author = {Pranjal Aggarwal and Sean Welleck},
year = {2026},
eprint = {2610.40565},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.40565},
}
Packages VLA policies, perception models and know-how as 250 reusable "embodied skills" and measures how much they help LLM agents on physical tasks.将 VLA 策略、感知模型与经验知识封装成 250 个可复用的“具身技能”，评测它们对大模型智能体完成物理任务的帮助。
@misc{han2026embodied,
title = {{Embodied Skill Bench: Benchmarking How Well Agents with Embodied Skills Perform on Physical Tasks}},
author = {Zhi Han and Zihan Guo and Chenxi Zeng and Zicai Cui and Yue Xin and Ming Zhou and Dongzhan Zhou and Weinan Zhang and Yang Li},
year = {2026},
eprint = {2610.16238},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.16238},
}
VLA robustness benchmark on RoboCasa365: paired clean/perturbed rollouts differing in one shift, plus stage-level diagnostics of where long-horizon failures start; evaluated on GR00T N1.5.基于 RoboCasa365 的 VLA 鲁棒性诊断基准：成对比较只差一个分布偏移因素的干净/扰动执行，并定位长时序任务中失败从哪个阶段开始，而不只看总成功率；在 GR00T N1.5 上评测。
@misc{wang2026robocasa,
title = {{RoboCasa-Shift: Stage-Aware Robustness Evaluation of Vision-Language-Action Policies in Long-Horizon Household Manipulation}},
author = {Haocheng Wang and Youquan Liu and Weiyu Guo and Aokai Zhang and Donghang Wu and Hongyu Liu and Sha Zhang and Weidong Yang and Ben Fei},
year = {2026},
eprint = {2610.27861},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.27861},
}
100 h of wearable-captured human loco-manipulation (mostly professional work) aligning egocentric video, whole-body and finger motion, touch and step-level text; reports 71.9% lower body error than baselines.用可穿戴设备采集的 100 小时人类移动操作数据，侧重职业场景，对齐第一视角视频、全身与手指动作、触觉和步骤级文字描述；其补全全身动作的方法报告身体关键点误差比最强基线低 71.9%。
@misc{zhou2026egoexplorer,
title = {{EgoExplorer: A Vision-Motion-Touch Observatory for Human Loco-Manipulation In Situ}},
author = {Yunsong Zhou and Xuekun Jiang and Haiqi Shen and Hangxu Liu and Chaojian Shi and Yujie Liu and Jingyuan Li and Haoran Yang and Jiafei Cao and Dongzhan Zhou and Dong Wang and Yao Mu and Chunhua Shen and Weinan Zhang},
year = {2026},
eprint = {2610.18872},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.18872},
}
Isaac Sim benchmark with 5 arms, 6 dexterous hands, ~200 assets and 18k human demos over 66 tasks; reports the best trained policy (a WAM) at only 29.1% on Core-50, near 0% on precision-fit and deformable tasks.基于 Isaac Sim 的灵巧操作基准：5 种机械臂、6 种灵巧手、约 200 个资产、1.8 万条人类演示、66 个任务；报告最好的训练策略（一种世界动作模型）在 Core-50 上仅 29.1%，精密装配和柔性物体任务接近 0%。
@misc{luo2026handson,
title = {{HandsOn: A Large-Scale Simulation Benchmark for Generalizable Dexterous Manipulation}},
author = {Tianwei Luo and Yiming He and Hongjin Wu and Ting Pan and Debin Ren and Haorui Guo and Junlin Yang and Yuxin Zuo and Zhehao Kou and Kaisen Yang and Zhuoqun Lai and Yi Ding and Jiale Yu and Yuchen Zhang and Zijie Qiu and Haozhan Li and Ning Ding},
year = {2026},
eprint = {2610.26765},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.26765},
}
Tests robot value (progress) models on 15 sim and 20 real tasks across task state, progress, failure/recovery and consistency; reports that good forward-progress scores hide weak history and recovery reasoning.在 15 个仿真和 20 个真实任务上评测机器人价值（进度）模型，覆盖任务状态、进度监测、失败与恢复、一致性四个维度；报告称前向进度得分高的模型在依赖历史和恢复判断上仍然薄弱。
@misc{liu2026robovalue,
title = {{RoboValue: A Fine-Grained Sim-and-Real Benchmark for Unified Evaluation of Robotic Value Models}},
author = {Shengbang Liu and Zhengye Du and Zhilong Wan and Honghao Su and Chenxiang Xia and Chang Ge and Jinyang Xiao and Nan Wang and Zhipeng Hu and Huining Yuan and Shu'ang Yu and Yue Chen and Tianxing Chen and Chuankang Li and Maoqing Yao and Wenbo Ding and Yu Wang and Guanghui Ren and Chao Yu},
year = {2026},
eprint = {2610.26663},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.26663},
}
216 LIBERO-based tasks with industrial constraints — conveyor loading/unloading/transfer, speed, time windows, tight tolerances; reports that VLAs and world action models degrade as constraints are combined.基于 LIBERO 的 216 个工业约束任务：传送带上料、下料与转运，可调节物体形状、带速、交互时间窗与公差；报告 VLA 和世界动作模型在多重约束叠加时性能逐步下降。
@misc{sung2026libero,
title = {{LIBERO-Industry: Benchmarking Robotic Manipulation under Industrial Process Constraints}},
author = {Ming Shin Cindy Sung and Duan Leijiahao and Haoxiang Wang and Boning Tan and Liangna Fu and Chao Tan and Yafu Li and Yulong Chen and Chaoqun Wan and Zhen Huang and Shuxian Liang and Xian-Sheng Hua},
year = {2026},
eprint = {2610.16782},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2610.16782},
}
The open data engine behind GR00T training: models that write timestamped action captions, dense process rewards and episode outcome rewards for human and robot video; reports SOTA on WGO-Bench, RoboReward and Robometer.GR00T 训练所用的开放数据引擎：一组模型为人类与机器人视频生成带时间戳的动作描述、稠密过程奖励和整段结果奖励；报告在 WGO-Bench、RoboReward 与 Robometer 上达到最优。
@misc{leng2026flora,
title = {{FLORA: An Open Model Suite for Scaling Robotics Data Engines}},
author = {Sicong Leng and Yunze Man and Zhiqi Li and Jing Wang and Yuhe Zhang and Ryan Punamiya and Siddharth Patki and Kaiyuan Zheng and Zi-ang Cao and Zhiding Yu and Sifei Liu and Danfei Xu and Ben Levin and Jan Kautz and Yuke Zhu and Johan Bjorck and Linxi "Jim" Fan},
year = {2026},
eprint = {2610.31779},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.31779},
}
106 sim tasks for two dexterous hands along five axes (generalization, instructions, few demos, long horizon, memory), each with an out-of-domain twin; 8,080 teleoperated demos; seven VLAs benchmarked.双灵巧手仿真基准：106 个任务按泛化、指令遵循、少示教、长时程、记忆五个维度组织，每个任务配有域外变体；提供 8,080 条遥操作示教，并在统一协议下评测了七个 VLA 模型。
@misc{wang2026bidex,
title = {{BiDex-100: A Comprehensive Benchmark for Bimanual Dexterous Manipulation Across 100+ Simulation Tasks}},
author = {Yong Wang and Kecheng Zheng and Xing Zhu and Chao Xu and Yujun Shen and Shengze Cai},
year = {2026},
eprint = {2610.25459},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.25459},
}
Benchmark for catching drifting, tumbling objects inside a spacecraft with a fixed-base arm, plus storage and servicing tasks. Reports a frozen generalist policy captured only 2/200 static targets.太空机器人操作基准：用固定基座机械臂抓取舱内漂浮、旋转的物体，另含样本存储与在轨维护任务。报告称所测的冻结通用策略在 200 个静止目标中只抓住 2 个。
@misc{wang2026astrobench,
title = {{AstroBench: Benchmarking Free-Floating Capture and Space Robotic Manipulation}},
author = {Zizhe Wang and Jia Deng and Xin Lu and Zehao Jia and Yi Huang and Yao Zhu},
year = {2026},
eprint = {2610.31064},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.31064},
}
Benchmark of 410 long-horizon task environments (manipulation, puzzles, household) for foundation-model robot agents, with a curated EP-100 split and a shared protocol checking goals, rules and physical validity.面向基础模型机器人智能体的长程任务基准：410 个任务环境（基础操作、序列谜题、多阶段家务），精选 EP-100 子集，统一协议同时检查目标完成、任务规则与物理有效性。
@misc{chen2026embodied,
title = {{Embodied Playground: A Long-Horizon Agentic Robotics Benchmark for Physical, Strategic, and Spatial Reasoning}},
author = {Feng Chen and Pei Zhou and Lixuan He and Yunchao Yao and Li Sun and Shenghua Gao and Mingyu Ding and Kaifeng Zhang and Kaichen Zhou and Yi Ma},
year = {2026},
eprint = {2610.27039},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.27039},
}
BEHAVIOR-Spatial, built on BEHAVIOR-1K: spatial QA, point grounding and robot action tests. A baseline policy drops from 41.6% on held-out layouts to 16.1% when the requested target changes; oracle target cues help most.基于 BEHAVIOR-1K 的 BEHAVIOR-Spatial 基准：包含空间问答、指点定位与机器人动作评测。基线策略在未见布局上成功率 41.6%，但在同一布局中更换目标物体指令时降至 16.1%；提供真值目标提示带来的提升最大。
@misc{chen2026behavior,
title = {{Behavior Spatial Benchmark: Evaluating Spatial Understanding for Robotics}},
author = {Sirui Chen and Huang Huang and Qineng Wang and Wenlong Huang and Manling Li and Ruohan Zhang and Li Fei-Fei and Jiajun Wu},
year = {2026},
eprint = {2610.34051},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.34051},
}
Benchmark where a robot must run physics experiments to find an unknown answer: 30 experiments, 5 disciplines, parameters resampled per instance. Reports even the best frontier agent succeeds under half the time.“为求知而行动”的基准：机器人在物理仿真实验室中自行操作仪器、读数，测出每次随机重采样的未知参数。共 30 个实验、5 个学科；报告最强的前沿模型智能体在原生控制下成功率也不到一半，失败多源于与仪器交互。
@misc{gao2026embodiedlab,
title = {{EmbodiedLab: Benchmarking Embodied Agents That Act to Know}},
author = {Hongcheng Gao and Zelin Zheng and Alex Zhou and Jay Zhu and Shijia Ge and Yazhe Wang and Xiping Li and Jianshu Zeng and Di Wu and Xuan Shangguan and Yexing Wan and Lingyu He and Yongli Wang and Xiaohan Yu and Zhiqi Jia and Yue Liu and Bryan Hooi and Yinpeng Dong and Wentao Zhang and Sihang Wu and Xiao He},
year = {2026},
eprint = {2610.35385},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.35385},
}
@misc{mao2026robotagentbench,
title = {{RobotAgentBench: Towards Open-world Robot Autonomy}},
author = {Lingjun Mao and Jinglin Cao and Lukun He and Wenpeng Xu and Yuchen Yan and Hengyu Jin and Zijun Zhang and Jinzhou Tang and Songyao Jin and Ziche Liu and Yifei Shao and Junbo Huang and Fang Nan and Ruobing Han and Ziqiao Xi and Ziming Xu and Shuang Liang and Sibo Zhu and Wenyi Wu and Xinyue Wang and Yiming Huang and Xinkai Zou and Gao Zhu and Kun Zhou and Biwei Huang},
year = {2026},
eprint = {2610.34649},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.34649},
}
Isaac Lab digital-twin benchmark for language-conditioned spatial tasks on the low-cost SO-101 arm (138 objects, 916 episodes); sim ranks policies exactly as the real arm. Best VLAs ~61%, an LLM-agent harness 83.2%.基于 Isaac Lab 数字孪生的低成本 SO-101 机械臂语言条件空间操作基准（138 个物体、916 个评测回合）；仿真中的策略排名与真机完全一致。最好的 VLA 约 61%，由 LLM 写代码调用抓取策略的方案达 83.2%。
Benchmark of 800 human-rated robot/hand videos from 10 generators to test MLLM judges on physical adherence and instruction alignment, plus a 12K-video training set and RoboJudge-9B (Pearson 0.719 with humans).评测多模态大模型能否评判具身视频生成：800 条人工标注视频（10 个生成模型，夹爪/人手/灵巧手），考察物理合理性与指令一致性；另提供 1.2 万条训练视频及 RoboJudge-9B（与人类评分 Pearson 0.719）。
@misc{ma2026robojudge,
title = {{RoboJudge: Multimodal Language Models as Judges for Embodied Video Generation}},
author = {Siyuan Ma and Yu Qi and Cheng Ye and Ruotian Peng and Xinyi Xu and Zitiantao Lin and Zedong Cai and Yijian Huang and Weiyang Liu and Siyuan Wang and Yandong Wen},
year = {2026},
eprint = {2610.27782},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2610.27782},
}
Benchmark of 22.5K verified egocentric interactions with pixel-precise contact points: current models often miss where to touch. As a memory bank for an affordance policy it lifts real grasp success from 16% to 58%.第一视角接触点预测基准：2.25 万个人工核验的交互事件，每个含接触前画面、语义意图与像素级接触位置。现有方法常常定位不准甚至找不到目标物体；把 EgoContact 用作已有可供性迁移策略的记忆库，真机抓取成功率从 16% 提升到 58%。
@misc{deryo2026egocontact,
title = {{EgoContact: Evaluating Physical Intelligence with Egocentric Video}},
author = {Adam Deryło and Alexey Gavryushin and Daniel Cremers and Marc Pollefeys and Xi Wang},
year = {2026},
eprint = {2610.39167},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.39167},
}
Benchmark built on a taxonomy of commonsense reasoning (occlusion, containment, affordances). Reports zero-shot policies dropping over 50% as distractors grow and acting on the wrong object 30% of the time in clutter.基于常识推理分类法（遮挡、容纳、支撑、可供性、隐藏物体）构建的操作基准。报告零样本通用策略在干扰物增多时成功率下降超过 50%，在杂乱场景中约 30% 的时间作用于错误物体，抓取正确部位的比例仅约一半。
@misc{guru2026molmospaces,
title = {{MolmoSpaces-CommonSense: Evaluating Commonsense in Robotic Manipulation Policies}},
author = {Arjun Guru and Max Argus and Jae Sung Park and Jiafei Duan and Ali Farhadi and Ranjay Krishna},
year = {2026},
eprint = {2610.41377},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.41377},
}
A language-conditioned drawing benchmark with open-ended valid solutions: 28,200 simulated demonstrations over 282 tasks, used to compare five VLA models on data scale, language grounding and multimodal behavior.语言条件绘画基准：同一指令有无数种合法画法。含 282 个任务的 28,200 条仿真示范，用于比较五种 VLA 模型在数据规模、语言落地和多模态行为上的差异。
@misc{xu2026burger,
title = {{BURGER: Benchmarking and Understanding Robotic Generalization via Embodied Drawing}},
author = {Xiangyu Xu and Bangshuai Peng and Ziyi Liu and Mengjie Zhao and Chenhao Lu and Zhengyang Fan and Huazhe Xu and Kaizhe Hu},
year = {2026},
eprint = {2610.40532},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.40532},
}
CoRL 2026 Oral. Measures diversity of imitation datasets directly from demonstrations, with entropy defined through a signature kernel, so no model has to be trained first.CoRL 2026 Oral。直接从示范数据度量模仿学习数据集的多样性：基于签名核定义熵，无需先训练模型。
@misc{sirigiri2026diversity,
title = {{Diversity You Can Actually Measure: A Fast, Model-Free Diversity Metric for Robotics Datasets}},
author = {Sreevardhan Sirigiri and Nathan Samuel de Lara and Christopher Agia and Florian Shkurti and Fabio Ramos},
year = {2026},
eprint = {2603.11634},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2603.11634},
}
@misc{jiang2026what,
title = {{What Are We Actually Benchmarking in Robot Manipulation?}},
author = {Tianchong Jiang and Xiangshan Tan and Samuel Wheeler and Luzhe Sun and Tewodros W. Ayalew and Matthew Walter},
year = {2026},
eprint = {2606.04233},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.04233},
}
CoRL 2026 Spotlight. A pipeline that turns egocentric human manipulation videos into robot training data, testing whether this helps VLA pretraining at scale.CoRL 2026 Spotlight。把第一视角人类操作视频转换为机器人训练数据的流程，并检验其能否大规模助力 VLA 预训练。
@misc{wang2026ego,
title = {{Ego2Robot: Scalable Robot Data Synthesis from Egocentric Human Data}},
author = {Ye Wang and Pei Lin and Xiong-Hui Chen and Haoqi Yuan and Zhixuan Liang and Yiyang Huang and Anzhe Chen and Zixing Lei and Jie Zhang and Tao Zhang and Haoyang Li and Tong Zhang and Chenxi Xiao and Ziyuan Jiao and Qin Jin},
year = {2026},
eprint = {2608.02580},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2608.02580},
}
CoRL 2026 Spotlight. Real-world benchmark that factorizes tabletop manipulation into motor and instruction atoms to test atomic skills and how policies recombine them.CoRL 2026 Spotlight。真实世界基准：把桌面操作分解为动作原子与指令原子，测试策略的原子技能及其重新组合能力。
@misc{wu2026atom,
title = {{ATOM-Bench: A Real-World Benchmark for Atomic Skills and Compositional Generalization in Manipulation Policies}},
author = {Zenan Wu and Bingqing Wei and Lu Liu and Zheqi He and Xi Wang and Jiakang Liu and Zehui Li and Guocai Yao and Jing-Shu Zheng and Xi Yang and Yongtao Wang},
year = {2026},
eprint = {2606.16826},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.16826},
}
An open-weight model that judges from video frames whether a manipulation attempt succeeded, useful for labeling episodes; reports 0.89 AUROC on unseen labs and robots.开放权重模型，根据视频帧判断操作是否成功，可用于标注数据；在未见过的实验室与机器人上报告 AUROC 0.89。
10 long-horizon and 30 stage-level bimanual dexterous tasks with physically grounded success conditions per stage, so partial progress is measurable; includes verified teleoperated demos.10 个长时序和 30 个阶段级双手灵巧任务，每个阶段都有物理判定的成功条件，可衡量中间进度；提供经过验证的遥操作示范和统一接口。
@misc{ping2026dexarena,
title = {{DexArena: A Unified Benchmark for Long-Horizon Dexterous Manipulation}},
author = {Haonan Ping and Jianshu Hu and Yimin Wang and Yi Zhu and Pai Liang and Qiyan Jin and Qirui Bao and Ying Qian and Xinyang Ding and Hongfei Qiao and Tianle Xue and Yucheng Bai and Soo Wei Jie and Shuguang Li and Yutong Ban},
year = {2026},
eprint = {2611.02629},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.02629},
}
Extends robot piano to one to five Shadow Hands, with robust, safe and multi-agent RL settings: at least 912 task settings and 36 baselines. Reports that top RL algorithms and LLM agents still struggle.把机器人弹钢琴扩展到 1 至 5 只 Shadow 手，并支持鲁棒、安全和多智能体强化学习：至少 912 种任务设置、36 个基线算法。作者报告前沿算法与大模型智能体在高难任务上仍很吃力。
@misc{xu2026omnipiano,
title = {{OmniPiano: Diverse Dexterous Piano-Playing Challenges for Standard, Robust, Safe, and Multi-Agent RL}},
author = {Fan Xu and Fan Yang and Yifan Xu and Ruiyuan Zhang and Yimou Wu and Nuo Yi and Muning Wen and Guang Chen and Costas Spanos and Shangding Gu},
year = {2026},
eprint = {2611.00243},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2611.00243},
}
Releases every episode of the ArmnetBench v0.1 evaluation on SO-101 arms: 600 teleoperated demos and 3,118 labelled policy rollouts, 73% of them failures, so many policies' mistakes can be compared on the same tasks.公开 ArmnetBench v0.1 在 SO-101 机械臂上的全部回合：600 条遥操作示范和 3118 条带标注的策略执行，其中 73% 是失败，可以在同一任务上比较多种策略的错误。
@misc{selvaraj2026rollout,
title = {{Rollout then Recycle: 3,718 Labelled Real-Robot Episodes from Benchmarking 7 Policies on 12 Tasks}},
author = {Praveen Selvaraj and Lorenzo Uttini and Ville Kuosmanen},
year = {2026},
eprint = {2611.03168},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.03168},
}
Sim platform where a fine-tuned VLA generates nearly 8,000 successful garment trajectories, re-rendered for RGB policies. Those beat policies trained on the original human data, but unseen garments stay hard.仿真平台：用微调后的 VLA 生成近 8000 条成功的衣物操作轨迹，并重新渲染供 RGB 策略训练，效果优于直接用人类数据训练，但泛化到未见衣物仍是难题。
@misc{huang2026clothdojo,
title = {{ClothDojo: Learned Data Generation and Benchmarking for Bimanual Garment Flattening and Folding}},
author = {Zilong Huang and Zipeng Ye and Caicheng Wang and Yuchen Xie and Jiaxiang Luo},
year = {2026},
eprint = {2611.01190},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.01190},
}
Argues VLA progress now depends more on data engines and evaluation than on architectures; finds a fidelity-versus-cost trade-off in datasets and gaps in compositional and long-horizon benchmarks.认为 VLA 的进展更取决于数据引擎和评测，而不是模型架构；指出数据集存在保真度与成本的权衡，评测基准在组合任务和长时程任务上仍有缺口。
@misc{wang2026vision,
title = {{Vision-Language-Action in Robotics: A Survey of Datasets, Benchmarks, and Data Engines}},
author = {Ziyao Wang and Bingying Wang and Hanrong Zhang and Tingting Du and Tianyang Chen and Guoheng Sun and Yexiao He and Zheyu Shen and Wanghao Ye and Ang Li},
year = {2026},
eprint = {2604.23001},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2604.23001},
}
Runs fourteen curricula under one protocol on five quadrupeds in MuJoCo MJX: no method tops every robot and level set, and methods with easy training levels lead on in-distribution random terrain.在 MuJoCo MJX 中用统一协议在五款四足上比较 14 种课程方法：没有方法在所有机器人和地形集上都领先，而训练关卡偏简单的方法在分布内随机地形上表现靠前。
@misc{gilewicz2026is,
title = {{Is It the Curriculum or Just the Easy Start? Benchmarking Unsupervised Environment Design on Quadrupeds}},
author = {Jakub Gilewicz and Dominik Galus and Piotr Wojdas},
year = {2026},
eprint = {2610.40364},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2610.40364},
}
A coding agent rebuilds a task in simulation from one calibration episode, then a video model renders photographic data; reports 93 vs. 84 of 100 real successes against teleop data, at about $107 per hour of data.编程智能体凭一条标定轨迹在仿真中重建任务，再由视频模型渲染成写实数据；报告真机 100 次成功 93 次，遥操作数据为 84 次，每小时数据约 107 美元。
@misc{zhang2026pyrua,
title = {{PyRUA-DataFarm: LLM Agent as a Cost-Efficient Teleoperator for Real-Ready Robot Data}},
author = {Kewei Zhang and Juncheng Ma and Guodong Lv and Duomin Wang and Feixiang Ruan and Ruiyang Si and Chuanrui Zhang and Yuxuan Chen and Hao Chen and Yichen Liu and Yintianrun Zhang and Yufan Deng and Enze Xie and Daquan Zhou},
year = {2026},
eprint = {2611.05320},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.05320},
}
Wraps each model once and each benchmark once, then schedules simulator workers and policy replicas as separate pools; reports evaluation 4.45x to 6.16x faster than the official StarVLA evaluator on one GPU.模型和基准各封装一次，把仿真 worker 与策略副本分成两个池调度；报告单卡评测比官方 StarVLA 评测器快 4.45 到 6.16 倍。
@misc{wu2026roboscope,
title = {{RoboScope Kit: Unified Interfaces for Fast and Scalable Embodied-AI Evaluation}},
author = {Deng-Ke Wu and Yifan Jin and Yida Yang and Lei Qian and Zhaoye Zhou and Hongru Zhang and Kai Ye and Shuyuan Hu and Yiran Mao and Zewei Ye and Bo Zhao},
year = {2026},
eprint = {2611.03850},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.03850},
}
A simulation benchmark for RL fine-tuning of VLAs without frequent resets; finds naive fine-tuning fails from object and end-effector shifts, and that targeted reset allocation and bias fixes mitigate it.面向无频繁重置场景下 VLA 强化学习微调的仿真基准；发现直接微调会因物体和末端位姿偏移而失败，定向分配重置并处理预训练偏差可以缓解。
A shared platform and dataset for human-data robot learning: 1,362 hours and 80k episodes across 1,965 tasks and 2,087 demonstrators, with contributions from labs and industry.面向人类数据机器人学习的共享平台和数据集：1,362 小时、8 万条片段，覆盖 1,965 项任务和 2,087 名演示者，接受学界和业界贡献。
@misc{punamiya2026egoverse,
title = {{EgoVerse: An Egocentric Human Dataset for Robot Learning from Around the World}},
author = {Ryan Punamiya and Simar Kareer and Zeyi Liu and Josh Citron and Ri-Zhao Qiu and Xiongyi Cai and Alexey Gavryushin and Jiaqi Chen and Davide Liconti and Lawrence Y. Zhu and Patcharapong Aphiwetsa and Baoyu Li and Aniketh Cheluva and Pranav Kuppili and Yangcen Liu and Dhruv Patel and Aidan Gao and Hye-Young Chung and Ryan Co and Renee Zbizika and Jeff Liu and Xiaomeng Xu and Haoyu Xiong and Geng Chen and Sebastiano Oliani and Wenkai Xuan and Chenyu Yang and Xi Wang and James Fort and Richard Newcombe and Josh Gao and Jason Chong and Garrett Matsuda and Aseem Doriwala and Marc Pollefeys and Robert Katzschmann and Xiaolong Wang and Shuran Song and Judy Hoffman and Danfei Xu},
year = {2026},
eprint = {2604.07607},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2604.07607},
}
Evaluates video world models by whether generated manipulation can be converted into actions and executed in physics simulation, not only whether it looks realistic.评估视频世界模型时不只看画面是否逼真，而是把生成的操作转成动作，在物理仿真中验证能否执行。
@misc{jiang2026robowm,
title = {{RoboWM-Bench: A Benchmark for Evaluating World Models in Robotic Manipulation}},
author = {Feng Jiang and Yang Chen and Kyle Xu and Yuchen Liu and Haifeng Wang and Zhenhao Shen and Jasper Lu and Shengze Huang and Yuanfei Wang and Chen Xie and Ruihai Wu},
year = {2026},
eprint = {2604.19092},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2604.19092},
}
Suites isolate generalisation across spatial layout, object instance, object category and subtask composition, plus bimanual BiMESA. The best of seven baselines reaches only 52% and 49%.五组评测分别隔离空间布局、物体实例、物体类别和子任务组合上的泛化能力，并提供双臂版 BiMESA。七种基线中最好的也只有 52% 和 49%，仍有很大提升空间。
@misc{wilcox2026mesa,
title = {{MESA: An Evaluation Framework for Compositional, Semantic, and Spatial Generalization in Robotics}},
author = {Albert Wilcox and Frank Chang and Nhi Nguyen and Aishani Chakraborty and Jeremy A. Collins and Vaibhav Saxena and Benjamin Joffe and Siddharth Karamcheti and Animesh Garg},
year = {2026},
eprint = {2611.05567},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.05567},
}
Sweeps one measured factor per scenario and certifies whether an agent should proceed or stop; finds the top learned CARLA agent at 0.60 decision accuracy against 0.74 for the rule-based expert.每类场景只改变一个可测因素，并认证智能体此时该通过还是该停下；CARLA 中得分最高的学习型智能体决策准确率为 0.60，规则专家为 0.74。
Benchmark showing VLA success often hides weak instruction following: scenes allow several valid tasks so language matters. Nine VLA and WAM policies drop sharply beyond the base level.揭示 VLA 的高成功率常掩盖薄弱的指令遵循：场景支持多种合法任务，必须依赖语言。九个 VLA 与 WAM 策略在基础级以外的测试中明显下降。
@misc{guo2026robofollow,
title = {{RoboFollow: Unveiling the Instruction Following Mirage in Embodied Agents}},
author = {Chang Guo and Yukun Xie and Bohan Tan and Zheng Chang and Zhaokai Yin and Qianli Ma and Yingqiao Wang and Chao Liang and Zhipeng Zhang},
year = {2026},
eprint = {2609.25636},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.25636},
}
Mobile-manipulation benchmark with 70 task scenarios combining navigation and manipulation in continuous spaces. Finds leading VLMs still struggle with spatial localization.移动操作评测基准：70 个在连续空间中结合导航与操作的任务场景。发现主流 VLM 在空间定位上仍有明显困难。
@misc{li2026rmmbench,
title = {{RMMBench: A Comprehensive Benchmark for Robotic Mobile Manipulation}},
author = {Huapeng Li and Fuxiang Feng and Jinqiu Fan and Shuo Yang and Fengjiao Chen and Xuezhi Cao and Ran Song and Wei Zhang},
year = {2026},
eprint = {2610.05414},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.05414},
}
Synthetic T-shirt folding and unfolding data: 6 robots, 1K shirts, 120K episodes. Policies trained only on it are reported above 90% success on unseen real shirts.合成 T 恤折叠与展开数据：6 种机器人、1000 件 T 恤、12 万条轨迹。报告仅用其训练的策略在未见过的真实 T 恤上成功率超过 90%。
@misc{chen2026foldnet,
title = {{FoldNet++: a Large-Scale Synthetic Dataset for Robotic T-Shirt Folding and Unfolding}},
author = {Yuxing Chen and Zhiyuan Wei and Bowen Xiao and Zhizheng Zhang and He Wang},
year = {2026},
eprint = {2609.12433},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.12433},
}
@misc{allshire2026scalable,
title = {{Scalable Behavior Cloning with Open Data, Training, and Evaluation}},
author = {Arthur Allshire and Himanshu Gaurav Singh and Ritvik Singh and Adam Rashid and Hongsuk Choi and David McAllister and Justin Yu and Yiyuan Chen and Huang Huang and Pieter Abbeel and Xi Chen and Rocky Duan and Phillip Isola and Jitendra Malik and Fred Shentu and Guanya Shi and Philipp Wu and Angjoo Kanazawa},
year = {2026},
eprint = {2606.27375},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.27375},
}
TouchSafeBench: 2,940 Habitat 3.0 episodes testing whether VLMs can tell if a robot is colliding or about to collide with people or the scene. Best average Macro-F1 stays below 50%.TouchSafeBench：基于 Habitat 3.0 的 2940 个场景，测试 VLM 能否判断机器人正在或即将与人或环境碰撞。最佳平均 Macro-F1 仍低于 50%。
Smartphone app for robot-free policy iteration: shows the current policy's predicted trajectory in AR so users collect corrective data where it fails. Reports up to 2x sample efficiency over offline demos.基于手机的免机器人策略迭代系统：用 AR 显示当前策略的预测轨迹，用户可在策略薄弱处采集纠正数据。报告样本效率最高为纯离线演示的 2 倍。
@misc{fang2026robopocket,
title = {{RoboPocket: Improve Robot Policies Instantly with Your Phone}},
author = {Junjie Fang and Wendi Chen and Han Xue and Fangyuan Zhou and Tian Le and Yi Wang and Yuting Zhang and Jun Lv and Chuan Wen and Cewu Lu},
year = {2026},
eprint = {2603.05504},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2603.05504},
}
HelpMeOut: 20,000 paired household activity episodes with nine neuropsychology-based error types; agents must spot when a person's task execution goes wrong. Learned, symbolic and LLM monitors all do poorly.HelpMeOut：2 万对家务活动片段，涵盖九类基于神经心理学的执行错误，要求智能体尽早发现人做事出错。学习型、符号型与大模型监控方法表现都较差。
@misc{wang2026helpmeout,
title = {{HelpMeOut: A Neuropsychology-Grounded Benchmark for Error-Aware Monitoring of Household Task Execution}},
author = {Haoran Wang and Jingwei Huang and Anqing Duan and Ho Seok Ahn and Bruce Macdonald},
year = {2026},
eprint = {2610.17694},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.17694},
}
RoboReel: a unified benchmark for policies learned from human videos, bundling real human demos, simulated robot trajectories and ten tasks with four test suites; sim trends hold on real-robot recreations.RoboReel：从人类视频学习策略的统一基准，打包真实人类演示视频、仿真机器人轨迹与十个任务及四套测试；在真机复现中仿真结论保持一致。
@misc{gu2026monkey,
title = {{Monkey See, Can Monkey Do? A Benchmark for Evaluating Robot Skill Learning by Observation}},
author = {Weiwei Gu and Anmol Gupta and Anant Sah and Ryan Varghese and Lalitha Shreya Vanam and Prabhath Adireddi and Peter Karkus and Nakul Gopalan},
year = {2026},
eprint = {2609.08209},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.08209},
}
600+ hours of whole-body human motion and object interaction from optical mocap with sub-millimetre marker accuracy, organised by FrameNet primitives, with a benchmark suite for humanoid learning.600 余小时的高精度全身人体运动与物体交互数据，由光学动捕采集（亚毫米级标记精度），按 FrameNet 动作基元组织，并配套面向人形机器人学习的评测套件。
@misc{ji2026hiphi,
title = {{HiPHI: A Large-Scale Benchmark for High-Precision Human Motion and Object-Interaction}},
author = {Jiahao Ji and Ji Ma and Runhan Zhang and Runyi Yu and Wenjia Wang and Weiheng Chi and Qianqian Peng and Weichao Yan and Yongfei Gu and Ye Tian and Ting Wu and Longwei Li and Chun Yuan and Ruoli Dai and Lei Han},
year = {2026},
eprint = {2608.16222},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2608.16222},
}
Four-level benchmark that widens the gap between a human demo and the robot's scene, plus IG-10K (20,000+ paired human-robot episodes). Nine models collapse at the hardest level; all stay below 13% zero-shot.四级基准逐步拉大人类演示与机器人场景的差异，并发布 IG-10K（2 万余条人-机配对片段）。九个模型在最高级别全面失效，未见任务零样本成功率均低于 13%。
@misc{zhou2026imitator,
title = {{The Imitator Game: Benchmarking Robot Imitative Ability Beyond Action Prediction}},
author = {Xunzhe Zhou and Yiyang Cai and Fengyi Wang and Ran Ju and Hanxiang Ren and Ruizhe Liu and Yu Zhang and Qian Luo and Feng Chen and Pei Zhou and Yi Ma and Yanchao Yang},
year = {2026},
eprint = {2608.22301},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2608.22301},
}
IMBench: 35 tasks and 14K trajectories testing reasoning and execution together under explicit constraints. VLMs reason partly but fail to plan executable steps; VLAs struggle with the constraints.IMBench：35 个任务、1.4 万条轨迹，在显式约束下同时考察物理推理与动作执行。VLM 有部分推理能力但给不出可执行计划，VLA 难以满足任务约束。
@misc{maurya2026imbench,
title = {{IMBench: A Benchmark for Intuitive Robotic Manipulation}},
author = {Anurag Maurya and Sukhvansh Jain and Prajwal Avhad and Gautham Balachandran and Ziyi Zhou and Atharva Kshirsagar and Satyam Singh and Bowen Li. Rishabh Mukund and Ritul Singh and Jatin Vira and Suvonil Chatterjee and Devesh K. Jha},
year = {2026},
eprint = {2607.15641},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2607.15641},
}
Reconstructs hand-object interaction from ordinary monocular human videos and retargets it into executable multi-fingered robot actions; beats prior work on hand-object estimation and gives a data-collection playbook.从普通单目人类视频重建手-物交互，并重定向为多指机器人手可执行的动作序列；手-物估计优于已有方法，并给出采集人类数据的实践建议。
@misc{paliwal2026do,
title = {{Do as I Do: Dexterous Manipulation Data from Everyday Human Videos}},
author = {Bhawna Paliwal and Haritheja Etukuru and William Liang and Pieter Abbeel and Nur Muhammad Mahi Shafiullah and Jitendra Malik},
year = {2026},
eprint = {2606.19333},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.19333},
}
Open-source data lifecycle layer for robot policy iteration: rollouts, reviews, datasets, training runs and evaluations become typed lineage artifacts that agents maintain, making iteration faster and auditable.开源的机器人策略迭代数据生命周期管理层：把执行记录、审核、数据集决策、训练与评测都表示为带类型的溯源工件，由智能体维护，让迭代更快、更可追溯。
@misc{luo2026robolineage,
title = {{RoboLineage: Agent-Native Data Lifecycle Governance Across Robot Policy Iterations}},
author = {Qian Luo and Wentao Guo and Zhennan Qin and Nanchun Guo and Yunhan Zhao and Yi Ma and Yanchao Yang},
year = {2026},
eprint = {2606.22142},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.22142},
}
Finger-driven handheld gripper for UMI-style data collection plus an 8,434-hour, 1.20M-episode, 119-task dataset; one policy transfers across UR, Franka and ELEY by mounting the gripper. Hardware and data released.手指驱动的手持夹爪，用于 UMI 式数据采集，并发布 8434 小时、120 万条、119 个任务的数据集；只需装上同款夹爪，同一策略即可迁移到 UR、Franka 与 ELEY。开源硬件与数据。
@misc{ohkawa2026yubi,
title = {{YUBI: Yielding Universal Bidigital Interface for Bimanual Dexterous Manipulation at Scale}},
author = {Takehiko Ohkawa and Jumpei Arima and Yuki Noguchi and Masatoshi Tateno and Makoto Sugiura and Takuya Okubo and Kengo Ikeuchi and Yuma Shin and Hiroki Nishizawa and Naoaki Kanazawa and Yuki Wakayama and Daiki Fukunaga and Koshi Makihara and Tomohiro Motoda and Floris Erich and Yukiyasu Domae and Tatsuya Matsushima and Yohishiro Okumatsu and Kei Ota},
year = {2026},
eprint = {2606.10244},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.10244},
}
55 scene-aligned 4D human motion sequences (150k+ frames) on rough terrain from egocentric wearables and a portable scanner, with an open pipeline; used to train perceptive locomotion run on a Unitree G1.利用第一人称可穿戴设备与便携式扫描仪采集 55 段与场景对齐的复杂地形 4D 人体运动（超过 15 万帧），并开源重建流程；用于训练感知式运动策略，部署到 Unitree G1。
@misc{brandes2026egohtr,
title = {{EgoHTR: Egocentric 4D Demonstrations of Human Terrain Traversal}},
author = {Alex Brandes and Haig Conti Georges Sajelian and Manthan Patel and Dominik Hollidt and Chenhao Li and Matthias Heyrman and Oliver Hausdoerfer and Manuel Kaufmann and Xi Wang and Jonas Frey and Angela P. Schoellig and Christian Holz and Marc Pollefeys and Marco Hutter},
year = {2026},
eprint = {2607.13472},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2607.13472},
}
Agentic generator of diverse, feasible manipulation tasks for data collection, with self-reflection and human-verified memory; 18k real trajectories collected. VLAs trained on its data generalize better.为数据采集自动生成多样且可执行的操作任务的智能体框架，结合自我反思与人工验证记忆；据此采集 1.8 万条真机轨迹，用其训练的 VLA 成功率与泛化更好。
@misc{zhang2026robogene,
title = {{RoboGene: Diversity-Driven Agentic Task Generation Framework for VLA Pre-training}},
author = {Yixue Zhang and Kun Wu and Zhi Gao and Zhen Zhao and Pei Ren and Zhiyuan Xu and Fei Liao and Xinhua Wang and Shichao Fan and Di Wu and Qiuxuan Feng and Meng Li and Zhengping Che and Chang Liu and Jian Tang},
year = {2026},
eprint = {2602.16444},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2602.16444},
}
@misc{wu2026navverse,
title = {{NavVerse: Benchmarking Indoor-to-Outdoor Embodied Navigation in Continuous Robot Simulation}},
author = {Junzhe Wu and Yue Hu and Zeyu Han and Po-Hsun Chang and Yinan Dong and Behrad Rabiei and Maani Ghaffari},
year = {2026},
eprint = {2607.19695},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2607.19695},
}
AutoDex runs real dexterous-grasp trials, labels outcomes and resets objects with little human help: 3,500+ trials on 100 objects (Allegro, Inspire); about 0.2 human-hours per 500 trajectories vs 9.3 by teleop.AutoDex 自动执行真机灵巧抓取、标注结果并复位物体，几乎无需人工：在 100 个物体上用 Allegro 与 Inspire 手采集 3500 多次试验；每 500 条轨迹约需 0.2 人时，遥操作需 9.3 人时。
@misc{choi2026autodex,
title = {{AutoDex: An Automated Real-World System for Dexterous Grasping Data Collection}},
author = {Mingi Choi and Gunhee Kim and Jisoo Kim and Taeksoo Kim and Taeyun Ha and Jongbin Lim and Hanbyul Joo},
year = {2026},
eprint = {2606.23689},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.23689},
}
Uses image editing of simulator renders as a middle step: edit in task semantics, recover 3D state from simulator priors, refine in physics, and turn it into robot supervision, e.g. dexterous grasps and goal states.以对仿真渲染图的图像编辑为中间表示：先编辑注入任务语义，再借助仿真先验恢复 3D 状态并在物理中细化，最后转为机器人可执行的监督，如灵巧抓取与目标状态。
@misc{wang2026im,
title = {{IM-ENGINE: Image Editing for Embodied Data Generation}},
author = {Yian Wang and Junyi Cao and Xiaowen Qiu and Chuang Gan},
year = {2026},
eprint = {2609.06279},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.06279},
}
Influence functions at billion-parameter VLA scale (about 313x faster via Kronecker structure and a rank-r Hessian approximation) for multi-task data curation; matches full-data fine-tuning with 50% of demos in sim.将影响函数扩展到十亿参数级 VLA（借助 Kronecker 结构与秩 r 的 Hessian 近似提速约 313 倍）用于多任务数据筛选；在仿真中只用 50% 示教即可达到全量数据微调的效果。
@misc{xu2026athena,
title = {{ATHENA: Accelerated Multi-Task Heterogeneous Influence Functions for Robot Data Curation}},
author = {Tao Xu and Jiaxin Wang and Runhao Zhang and Jiayi Guan and Xianchao Zeng and Weixi Song and Xinyu Zhou and Zhetao Chen and Guang Chen and Yong-Lu Li},
year = {2026},
eprint = {2606.16208},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.16208},
}
Real-to-sim benchmark for human-to-robot transfer: real human video demos plus simulated robot demos on four tasks, with a fixed protocol to compare how H2R methods scale with more human data.评估“人到机器人”迁移的 Real2Sim 基准：四个任务配有真实人类视频示教和仿真机器人示教，以统一协议比较各方法随人类数据增加的扩展能力。
@misc{xiao2026h,
title = {{H2RBench: A Real-to-Sim Benchmark for Evaluating Human-to-Robot Transfer}},
author = {Chuyang Xiao and Haotian Zhan and Sriram Krishna and Peilin Meng and Muhammad Zubair Irshad and Sergey Zakharov and David Held},
year = {2026},
eprint = {2609.24778},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.24778},
}
Auto-labels robot demos with boxes, object trajectories and phase labels, plus a reliability score from the task's spatio-temporal structure, so fewer noisy labels are kept and fewer good ones are thrown out.为机器人示教自动标注边界框、物体轨迹与操作阶段，并利用任务的时空结构为每个标注给出可靠性分数，从而少留噪声标签、少丢有用样本。
@misc{blank2026sparc,
title = {{SPARC: Reliable Spatial Annotations from Robot Demonstrations at Scale}},
author = {Nils Blank and Paul Mattes and Maximilian Xiling Li and Jakub Suliga and Thomas Roth and Moritz Reuss and Pankhuri Vanjani and Rudolf Lioutikov},
year = {2026},
eprint = {2606.13497},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.13497},
}
An imitation-learning benchmark on the open KlaskTron table game robot: 80k real human demo steps, augmented and synthetic data, and baselines exposing BC and GAIL failure modes; policies run on hardware unchanged.基于开源 KlaskTron 桌面对抗游戏机器人的模仿学习基准：8 万步真实人类示教、对称增强与合成数据，以及揭示 BC 和 GAIL 失效模式的基线；训练出的策略可直接上真机运行。
Human demos of creative tool use (repurposing objects when the usual tool is missing) from 1 ego + 8 exo cameras with hand, tool and object labels; benchmarks reasoning, 3D/4D reconstruction and policy learning.记录人类创造性使用工具（缺少常规工具时改用其他物体）的示范，含一路第一视角和八路第三视角相机及手、工具、物体标注；基准涵盖推理、三维/四维重建与策略学习。
@misc{elmoghany2026tooldex,
title = {{ToolDex: An Ego-Exo Human Demonstration Dataset for Creative Physical Tool Use}},
author = {Mohamed Elmoghany and Kehan Li and Ruoqu Chen and Tarun Chiruvolu and Shutong Zhang and Ryan Wang and Stephen Tian and Yaru Niu and Michelle Guo and Ding Zhao and Weiyu Liu and Mengdi Xu and Li Fei-Fei and Jiajun Wu},
year = {2026},
eprint = {2611.09036},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.09036},
}
Agents revise their own harness under a fixed budget on vision and robotics tasks; self-improvement helps most when the agent builds its own solution; open-weight models lag most under strict physical checks.评测智能体在固定预算下改写自身框架的能力，任务涵盖视觉与机器人；发现智能体自建解法时自我改进收益最大，开源权重模型在需通过物理稳定性等严格检查的任务上落后最多。
@misc{zhang2026flex,
title = {{Flex-RSI: Benchmarking Recursive Self-Improvement of Frontier Agents}},
author = {Jingdong Zhang and Jionghao Wang and Zhengming Yu and Yancheng Cai and Wei Chow and Chris Choy and Zhiyang Dou and Hang Hua and Jiachen Li and Zhixuan Liu and Chen Wang and Songlin Wei and Siyuan Yang and Yu Yuan and Xiaohang Zhan and Yucheng Zhang and Shuhong Zheng and Xin Li and Lingjie Liu and Wojciech Matusik and Zhengzhong Tu and Wenping Wang},
year = {2026},
eprint = {2611.08946},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2611.08946},
}
Manipulation benchmark where simulated humans cue goals, finish parts of the task, interrupt or share the workspace. Reports that VLA and IL policies drop when humans act, and higher success does not mean safer motion.让仿真人类参与其中的操作基准：人会提示目标、完成部分任务、打断或与机器人共处同一工作区。报告显示 VLA 与模仿学习策略在有人活动时性能下降，且成功率更高并不意味着更安全。
@misc{zhou2026harp,
title = {{HARP-Bench: Benchmarking Robot Manipulation Around Active Humans}},
author = {Qinhong Zhou and Riley Zong and Chunru Lin and Diyi Hu and Shivraj Singh Bhatti and Hongxin Zhang and Lin Qiu and Hanqing Zeng and Chuang Gan},
year = {2026},
eprint = {2611.11123},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.11123},
}
200+ hours from 14 people: six head cameras, wrist vision and IMU, plus 120 Hz optical full-body mocap, covering walking, terrain and carrying. Benchmarks body reconstruction, foot contact and humanoid retargeting.14 名参与者、200 多小时数据：六个头戴相机、腕部视觉与惯性传感，加 120 Hz 光学全身动捕，覆盖行走、地形穿越与搬运等。提供全身重建、足部接触估计与人形重定向三项基准。
@misc{deshmukh2026robomocap,
title = {{RoboMoCap: A Large-Scale Whole-Body Motion Capture Dataset with Egocentric Vision for Robot Learning}},
author = {Mayur Deshmukh and Aaron Tung and Junyi Li and Nikash Bhardwaj and Ademi Adeniji and Vincent Liu and Michael Cho and Jiafei Duan},
year = {2026},
eprint = {2611.10994},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.10994},
}
'Stochastic challenges': reproducible real scenes with tasks sampled from a shared distribution, so robots in different labs compare by success rate like a hidden test set. Shown with two robots in two labs.提出“随机挑战”评测范式：可复现的真实场景 + 从同一分布随机抽取任务，让不同实验室的机器人能像隐藏测试集一样按成功率比较。在一个移动操作基准上用两个实验室的两台机器人做了演示。
@misc{glaer2026benchmarking,
title = {{Benchmarking Real-World Robotic Manipulation through Stochastic Challenges}},
author = {Elias Gläßer and Gülhan Şikaroğlu and Pavlo Kupyn and Marc Pollefeys and Zuria Bauer and Hermann Blum},
year = {2026},
eprint = {2611.12804},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.12804},
}
A portable hand-held rig for collecting demonstrations with a high-DoF robot hand, using the same hand hardware for collection and deployment, so there is no retargeting from human hands and no full robot setup needed.便携手持式数据采集装置，面向高自由度机器人灵巧手；采集与部署使用同一只机械手，无需从人手动作重定向，也无需搭建完整机器人系统。
@misc{xie2026dexport,
title = {{DexPort: A Hand-Held Skill Interface for Dexterous Hand Data Collection}},
author = {Amber Xie and Jiawei Fu and Xiaofeng Guo and Carmelo Sferrazza and Guanya Shi and Haozhi Qi},
year = {2026},
eprint = {2610.32969},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.32969},
}
Fully open UMI-style handheld gripper with a low-cost stereo camera for metric, low-drift trajectories, plus UMI-Loc-Bench-v1 for localization accuracy; hardware, software and data under a permissive license.完全开源的 UMI 式手持夹爪：3D 打印、标准 ISO 接口，用低成本双目相机获得公制尺度、低漂移的轨迹，并提出手持采集设备定位精度基准 UMI-Loc-Bench-v1；硬件、软件与数据以可商用许可证发布。
@misc{murali2026openjaws,
title = {{OpenJaws: An Open Hardware UMI Framework with Stereo Localization for Generalist Manipulation}},
author = {Adithyavairavan Murali and Yash Oza and Karanbir Chahal and Neel Jawale and Silas De Vries and Hesam Rabeti and Erwin Coumans and Balakumar Sundaralingam and Stan Birchfield},
year = {2026},
eprint = {2610.36598},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.36598},
}
Open-source bimanual UMI with camera, LiDAR-inertial sensing, compute and power on board: no external tracking or calibration; parallel SLAM outputs training-ready data. Releases 40 h of data and hardware files.全栈开源的双手 UMI 野外数据采集系统：手持设备集成相机、激光雷达-惯性传感、板载算力与电源，无需外部追踪或现场标定；异步 SLAM 流水线直接产出可训练数据集，并发布 40 小时数据、硬件设计与复现指南，用 π0.5 和 MolmoAct2 微调验证数据质量。
@misc{velamoor2026umi,
title = {{UMI-Live for In-The-Wild Bimanual Robotics Data Collection}},
author = {Govind Velamoor and Adrian Lin and Sam Jung and Spencer Mateega and Brandon Y. Yang and Amish Sethi},
year = {2026},
eprint = {2610.39385},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.39385},
}
~500 h / 100K trajectories on one mobile bimanual robot (R1 Lite) across 50 real scenes, plus G0: a VLM planner over a flow-matching VLA; reports that single-embodiment pretraining matters most for few-shot transfer.在单一移动双臂机器人（R1 Lite）上采集约 500 小时、10 万条轨迹、覆盖 50 个真实场景，并提出 G0：VLM 规划器 + 流匹配 VLA 执行器的双系统模型；论文报告单本体预训练对少样本迁移最关键。
@misc{jiang2025galaxea,
title = {{Galaxea Open-World Dataset and G0 Dual-System VLA Model}},
author = {Tao Jiang and Tianyuan Yuan and Yicheng Liu and Chenhao Lu and Jianning Cui and Xiao Liu and Shuiqi Cheng and Jiyang Gao and Huazhe Xu and Hang Zhao},
year = {2025},
eprint = {2509.00576},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2509.00576},
}
Isaac Sim platform where an LLM builds tasks and layouts from a task-oriented scene graph over ~10K objects; its 200-scenario benchmark finds modular foundation-model systems generalize better than end-to-end policies.基于 Isaac Sim 的平台：LLM 依据任务导向场景图，从约 1 万个物体中生成任务与布局；其 200 个场景的基准发现，结合基础模型的模块化系统比端到端策略泛化更好。
@misc{gao2025genmanip,
title = {{GenManip: LLM-driven Simulation for Generalizable Instruction-Following Manipulation}},
author = {Ning Gao and Yilun Chen and Shuai Yang and Xinyi Chen and Yang Tian and Hao Li and Haifeng Huang and Hanqing Wang and Tai Wang and Jiangmiao Pang},
year = {2025},
eprint = {2506.10966},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2506.10966},
}
A million-scale real-robot dataset collected in purpose-built scenes, with the GO-1 generalist policy trained on it.在专门搭建的场景中采集的百万级真实机器人数据集，并基于其训练通用策略 GO-1。
@misc{agibotworldcontributors2025agibot,
title = {{AgiBot World Colosseo: A Large-scale Manipulation Platform for Scalable and Intelligent Embodied Systems}},
author = {AgiBot-World-Contributors and Qingwen Bu and Jisong Cai and Li Chen and Xiuqi Cui and Yan Ding and Siyuan Feng and Shenyuan Gao and Xindong He and Xuan Hu and Xu Huang and Shu Jiang and Yuxin Jiang and Cheng Jing and Hongyang Li and Jialu Li and Chiming Liu and Yi Liu and Yuxiang Lu and Jianlan Luo and Ping Luo and Yao Mu and Yuehan Niu and Yixuan Pan and Jiangmiao Pang and Yu Qiao and Guanghui Ren and Cheng Ruan and Jiaqi Shan and Yongjian Shen and Chengshi Shi and Mingkang Shi and Modi Shi and Chonghao Sima and Jianheng Song and Huijie Wang and Wenhao Wang and Dafeng Wei and Chengen Xie and Guo Xu and Junchi Yan and Cunbiao Yang and Lei Yang and Shukai Yang and Maoqing Yao and Jia Zeng and Chi Zhang and Qinglin Zhang and Bin Zhao and Chengyue Zhao and others},
year = {2025},
eprint = {2503.06669},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2503.06669},
}
829 hours of egocentric Apple Vision Pro video with tracked 3D poses for every hand joint, across 194 tabletop tasks from tying shoelaces to folding laundry; includes hand-trajectory prediction benchmarks.用 Apple Vision Pro 采集的 829 小时第一视角视频，配有每个手部关节的三维姿态，覆盖从系鞋带到叠衣服的 194 种桌面任务，并提供手部轨迹预测基准。
@misc{hoque2025egodex,
title = {{EgoDex: Learning Dexterous Manipulation from Large-Scale Egocentric Video}},
author = {Ryan Hoque and Peide Huang and David J. Yoon and Mouli Sivapurapu and Jian Zhang},
year = {2025},
eprint = {2505.11709},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2505.11709},
}
CoRL 2026 Spotlight. Humanoid locomotion dataset from internet video that targets the floating, penetration and foot skating common in earlier video-derived data.CoRL 2026 Spotlight。源自网络视频的人形运动数据集，针对此前视频数据常见的悬空、穿透与脚滑问题。
@misc{lee2025phuma,
title = {{PHUMA: Physically Reliable Humanoid Locomotion Dataset}},
author = {Kyungmin Lee and Sibeen Kim and Youngdo Lee and Minho Park and Hyunseung Kim and Dongyoon Hwang and Donghu Kim and Hojoon Lee and Jaegul Choo},
year = {2025},
eprint = {2510.26236},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2510.26236},
}
Report on an online system for testing many robot policies on many real tasks, plus a study of recent VLAs on its first benchmark, Table30.介绍一套在线系统，用于在真机上大规模评测多种机器人策略，并报告近期 VLA 在首个基准 Table30 上的表现。
@misc{yakefu2025robochallenge,
title = {{RoboChallenge: Large-scale Real-robot Evaluation of Embodied Policies}},
author = {Adina Yakefu and Bin Xie and Chongyang Xu and Enwen Zhang and Erjin Zhou and Fan Jia and Haitao Yang and Haoqiang Fan and Haowei Zhang and Hongyang Peng and Jing Tan and Junwen Huang and Kai Liu and Kaixin Liu and Kefan Gu and Qinglun Zhang and Ruitao Zhang and Saike Huang and Shen Cheng and Shuaicheng Liu and Tiancai Wang and Tiezhen Wang and Wei Sun and Wenbin Tang and Yajun Wei and Yang Chen and Youqiang Gui and Yucheng Zhao and Yunchao Ma and Yunfei Wei and Yunhuan Yang and Yutong Guo and Ze Chen and Zhengyuan Du and Ziheng Zhang and Ziming Liu and Ziwei Yan},
year = {2025},
eprint = {2510.17950},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2510.17950},
}
Portable, mocap-free humanoid teleoperation and data collection built around a VR headset, aimed at making whole-body demonstrations scale.基于 VR 头显的便携式免动捕人形遥操作与数据采集系统，目标是让全身示教数据能够规模化采集。
@misc{ze2025twist,
title = {{TWIST2: Scalable, Portable, and Holistic Humanoid Data Collection System}},
author = {Yanjie Ze and Siheng Zhao and Weizhuo Wang and Angjoo Kanazawa and Rocky Duan and Pieter Abbeel and Guanya Shi and Jiajun Wu and C. Karen Liu},
year = {2025},
eprint = {2511.02832},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2511.02832},
}
ASIMOV v2: physical-safety benchmark built from real injury narratives turned into generated images and videos, testing whether foundation models notice danger and intervene, plus safety-reasoning post-training.ASIMOV v2：把真实伤害事故描述转成生成的图像与视频，构建可持续扩展的物理安全基准，测试基础模型能否察觉危险并干预；并提出安全推理后训练方法。
@misc{jindal2025can,
title = {{Can AI Perceive Physical Danger and Intervene?}},
author = {Abhishek Jindal and Dmitry Kalashnikov and R. Alex Hofer and Oscar Chang and Divya Garikapati and Anirudha Majumdar and Pierre Sermanet and Vikas Sindhwani},
year = {2025},
eprint = {2509.21651},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2509.21651},
}
TVL pairs tactile and visual observations with language descriptions and trains a touch-vision-language encoder plus a model that describes how things feel.TVL 数据集将触觉与视觉观测配上语言描述，并训练触觉-视觉-语言编码器以及能描述物体触感的模型。
@misc{fu2024touch,
title = {{A Touch, Vision, and Language Dataset for Multimodal Alignment}},
author = {Letian Fu and Gaurav Datta and Huang Huang and William Chung-Ho Panitch and Jaimyn Drake and Joseph Ortiz and Mustafa Mukadam and Mike Lambeta and Roberto Calandra and Ken Goldberg},
year = {2024},
eprint = {2402.13232},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2402.13232},
}
Extends MimicGen to humanoids with two dexterous hands, turning 60 human demos into 21K simulated ones across bimanual tasks, plus a real-to-sim-to-real pipeline tested on real can sorting.把 MimicGen 扩展到双灵巧手人形机器人，从 60 条人类演示生成 2.1 万条双臂仿真演示，并给出在真实分拣罐子任务上验证的 real-to-sim-to-real 流程。
@misc{jiang2024dexmimicgen,
title = {{DexMimicGen: Automated Data Generation for Bimanual Dexterous Manipulation via Imitation Learning}},
author = {Zhenyu Jiang and Yuqi Xie and Kevin Lin and Zhenjia Xu and Weikang Wan and Ajay Mandlekar and Linxi Fan and Yuke Zhu},
year = {2024},
eprint = {2410.24185},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2410.24185},
}
Benchmark of 40 home tasks for mobile bimanual robots, from reaching to kitchen cleaning, each with human-collected demonstrations.面向移动双臂机器人的基准，含 40 个家庭任务（从抓取到厨房清洁），均配人类采集的演示。
@misc{chernyadev2024bigym,
title = {{BiGym: A Demo-Driven Mobile Bi-Manual Manipulation Benchmark}},
author = {Nikita Chernyadev and Nicholas Backshall and Xiao Ma and Yunfan Lu and Younggyo Seo and Stephen James},
year = {2024},
eprint = {2407.07788},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2407.07788},
}
GPU-accelerated Home Assistant Benchmark with realistic low-level control; the authors report over 3x the speed of prior magical-grasp versions.GPU 加速的 Home Assistant Benchmark，采用真实低层控制；作者报告速度超过此前“魔法抓取”实现的 3 倍。
@misc{shukla2024maniskill,
title = {{ManiSkill-HAB: A Benchmark for Low-Level Manipulation in Home Rearrangement Tasks}},
author = {Arth Shukla and Stone Tao and Hao Su},
year = {2024},
eprint = {2412.13211},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2412.13211},
}
@misc{khazatsky2024droid,
title = {{DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset}},
author = {Alexander Khazatsky and Karl Pertsch and Suraj Nair and Ashwin Balakrishna and Sudeep Dasari and Siddharth Karamcheti and Soroush Nasiriany and Mohan Kumar Srirama and Lawrence Yunliang Chen and Kirsty Ellis and Peter David Fagan and Joey Hejna and Masha Itkina and Marion Lepert and Yecheng Jason Ma and Patrick Tree Miller and Jimmy Wu and Suneel Belkhale and Shivin Dass and Huy Ha and Arhan Jain and Abraham Lee and Youngwoon Lee and Marius Memmel and Sungjae Park and Ilija Radosavovic and Kaiyuan Wang and Albert Zhan and Kevin Black and Cheng Chi and Kyle Beltran Hatch and Shan Lin and Jingpei Lu and Jean Mercat and Abdul Rehman and Pannag R Sanketi and Archit Sharma and Cody Simpson and Quan Vuong and Homer Rich Walke and Blake Wulfe and Ted Xiao and Jonathan Heewon Yang and Arefeh Yavary and Tony Z. Zhao and Christopher Agia and Rohan Baijal and Mateo Guaman Castro and Daphne Chen and Qiuyu Chen and others},
year = {2024},
eprint = {2403.12945},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2403.12945},
}
Large real-robot teleoperation dataset across several embodiments (single-arm, dual-arm, humanoid) collected under one standardized protocol, and it keeps failed demos with annotated causes.在统一规范下采集的大规模真机遥操作数据集，覆盖单臂、双臂与人形等多种本体，并保留了标注失败原因的失败演示。
@misc{wu2024robomind,
title = {{RoboMIND: Benchmark on Multi-embodiment Intelligence Normative Data for Robot Manipulation}},
author = {Kun Wu and Chengkai Hou and Jiaming Liu and Zhengping Che and Xiaozhu Ju and Zhuqin Yang and Meng Li and Yinuo Zhao and Zhiyuan Xu and Guang Yang and Shichao Fan and Xinhua Wang and Fei Liao and Zhen Zhao and Guangyu Li and Zhao Jin and Lecheng Wang and Jilei Mao and Ning Liu and Pei Ren and Qiang Zhang and Yaoxu Lyu and Mengzhen Liu and Jingyang He and Yulin Luo and Zeyu Gao and Chenxuan Li and Chenyang Gu and Yankai Fu and Di Wu and Xingyu Wang and Sixiang Chen and Zhenyu Wang and Pengju An and Siyuan Qian and Shanghang Zhang and Jian Tang},
year = {2024},
eprint = {2412.13877},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2412.13877},
}
27 simulated tasks (12 locomotion, 15 whole-body manipulation) on a Unitree H1 with two dexterous hands; reports that DreamerV3, TD-MPC2, SAC and PPO fail most of them, especially long-horizon whole-body ones.27 个仿真任务（12 个运动、15 个全身操作），机器人为带两只灵巧手的 Unitree H1；报告 DreamerV3、TD-MPC2、SAC、PPO 在大多数任务上未达标，长时程全身协调任务尤甚。
@misc{sferrazza2024humanoidbench,
title = {{HumanoidBench: Simulated Humanoid Benchmark for Whole-Body Locomotion and Manipulation}},
author = {Carmelo Sferrazza and Dun-Ming Huang and Xingyu Lin and Youngwoon Lee and Pieter Abbeel},
year = {2024},
eprint = {2403.10506},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2403.10506},
}
@misc{pumacay2024colosseum,
title = {{THE COLOSSEUM: A Benchmark for Evaluating Generalization for Robotic Manipulation}},
author = {Wilbert Pumacay and Ishika Singh and Jiafei Duan and Ranjay Krishna and Jesse Thomason and Dieter Fox},
year = {2024},
eprint = {2402.08191},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2402.08191},
}
100 task categories over 2,000+ objects that need common sense, world knowledge and multi-step reasoning, not just instruction following; evaluates VLAs, embodied agents and VLMs in one suite.100 类任务、2000 多个物体，任务需要常识、世界知识与多步推理，而不只是照指令执行；在同一套件中评测 VLA、具身智能体与 VLM。
@misc{zhang2024vlabench,
title = {{VLABench: A Large-Scale Benchmark for Language-Conditioned Robotics Manipulation with Long-Horizon Reasoning Tasks}},
author = {Shiduo Zhang and Zhe Xu and Peiju Liu and Xiaopeng Yu and Yuan Li and Qinghui Gao and Zhaoye Fei and Zhangyue Yin and Zuxuan Wu and Yu-Gang Jiang and Xipeng Qiu},
year = {2024},
eprint = {2412.18194},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2412.18194},
}
@misc{zhu2024point,
title = {{Point Cloud Matters: Rethinking the Impact of Different Observation Spaces on Robot Learning}},
author = {Haoyi Zhu and Yating Wang and Di Huang and Weicai Ye and Wanli Ouyang and Tong He},
year = {2024},
eprint = {2402.02500},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2402.02500},
}
Turns a handful of human demos into thousands by splitting them into object-centric segments and replaying those segments in new scenes; widely reused to scale imitation data.把少量人类演示切分为以物体为中心的片段，并在新场景中变换重放，从而扩增为数千条演示；被广泛用于扩充模仿学习数据。
@misc{mandlekar2023mimicgen,
title = {{MimicGen: A Data Generation System for Scalable Robot Learning using Human Demonstrations}},
author = {Ajay Mandlekar and Soroush Nasiriany and Bowen Wen and Iretiayo Akinola and Yashraj Narang and Linxi Fan and Yuke Zhu and Dieter Fox},
year = {2023},
eprint = {2310.17596},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2310.17596},
}
@misc{walke2023bridgedata,
title = {{BridgeData V2: A Dataset for Robot Learning at Scale}},
author = {Homer Walke and Kevin Black and Abraham Lee and Moo Jin Kim and Max Du and Chongyi Zheng and Tony Zhao and Philippe Hansen-Estruch and Quan Vuong and Andre He and Vivek Myers and Kuan Fang and Chelsea Finn and Sergey Levine},
year = {2023},
eprint = {2308.12952},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2308.12952},
}
@misc{fang2023rh,
title = {{RH20T: A Comprehensive Robotic Dataset for Learning Diverse Skills in One-Shot}},
author = {Hao-Shu Fang and Hongjie Fang and Zhenyu Tang and Jirong Liu and Chenxi Wang and Junbo Wang and Haoyi Zhu and Cewu Lu},
year = {2023},
eprint = {2307.00595},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2307.00595},
}
Time-synced first-person and third-person video of people performing skilled activities (cooking, sports, repair...), with rich annotations; a large source for learning manipulation skills from humans.人们进行技能活动（烹饪、运动、维修等）时同步采集的第一人称与第三人称视频，附丰富标注，是从人类视频学习操作技能的重要数据来源。
@misc{grauman2023ego,
title = {{Ego-Exo4D: Understanding Skilled Human Activity from First- and Third-Person Perspectives}},
author = {Kristen Grauman and Andrew Westbury and Lorenzo Torresani and Kris Kitani and Jitendra Malik and Triantafyllos Afouras and Kumar Ashutosh and Vijay Baiyya and Siddhant Bansal and Bikram Boote and Eugene Byrne and Zach Chavis and Joya Chen and Feng Cheng and Fu-Jen Chu and Sean Crane and Avijit Dasgupta and Jing Dong and Maria Escobar and Cristhian Forigua and Abrham Gebreselasie and Sanjay Haresh and Jing Huang and Md Mohaiminul Islam and Suyog Jain and Rawal Khirodkar and Devansh Kukreja and Kevin J Liang and Jia-Wei Liu and Sagnik Majumder and Yongsen Mao and Miguel Martin and Effrosyni Mavroudi and Tushar Nagarajan and Francesco Ragusa and Santhosh Kumar Ramakrishnan and Luigi Seminara and Arjun Somayazulu and Yale Song and Shan Su and Zihui Xue and Edward Zhang and Jinxu Zhang and Angela Castillo and Changan Chen and Xinzhu Fu and Ryosuke Furuta and Cristina Gonzalez and Prince Gupta and Jiabo Hu and others},
year = {2023},
eprint = {2311.18259},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2311.18259},
}
Benchmark for language-grounded tasks with continuous object goal states in realistic 3D scenes, where most suites assume binary goals.面向语言任务的基准，目标是连续物体状态（多数基准只有二值目标），场景为真实感三维环境。
@misc{gong2023arnold,
title = {{ARNOLD: A Benchmark for Language-Grounded Task Learning With Continuous States in Realistic 3D Scenes}},
author = {Ran Gong and Jiangyong Huang and Yizhou Zhao and Haoran Geng and Xiaofeng Gao and Qingyang Wu and Wensi Ai and Ziheng Zhou and Demetri Terzopoulos and Song-Chun Zhu and Baoxiong Jia and Siyuan Huang},
year = {2023},
eprint = {2304.04321},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2304.04321},
}
Over a million physically validated ShadowHand grasps on thousands of objects, synthesized by optimizing a differentiable force-closure energy; a standard training set for dexterous grasp models.通过优化可微的力封闭能量合成、并经物理仿真验证的百万级 ShadowHand 抓取，覆盖数千个物体，是灵巧抓取模型的常用训练集。
@misc{wang2022dexgraspnet,
title = {{DexGraspNet: A Large-Scale Robotic Dexterous Grasp Dataset for General Objects Based on Simulation}},
author = {Ruicheng Wang and Jialiang Zhang and Jiayi Chen and Yinzhen Xu and Puhao Li and Tengyu Liu and He Wang},
year = {2022},
eprint = {2210.02697},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2210.02697},
}
800K+ annotated 3D object models; widely used to populate simulators with diverse objects for grasping, rearrangement and procedural scene generation (Objaverse-XL extends it to 10M+).80 万+ 带标注的 3D 物体模型，常用于为仿真器填充多样物体，用于抓取、重排和程序化场景生成（Objaverse-XL 扩展到 1000 万+）。
@misc{deitke2022objaverse,
title = {{Objaverse: A Universe of Annotated 3D Objects}},
author = {Matt Deitke and Dustin Schwenk and Jordi Salvador and Luca Weihs and Oscar Michel and Eli VanderBilt and Ludwig Schmidt and Kiana Ehsani and Aniruddha Kembhavi and Ali Farhadi},
year = {2022},
eprint = {2212.08051},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2212.08051},
}
2.4M egocentric RGB-D frames (4,000+ sequences, 800 object instances in 16 categories, 610 rooms) of people handling rigid and articulated objects, with segmentation, hand and object pose labels.240 万帧第一视角 RGB-D（4000 多段序列、16 类 800 个物体实例、610 个房间），记录人操作刚体与铰接物体，带分割、手部与物体位姿标注。
@misc{liu2022hoi,
title = {{HOI4D: A 4D Egocentric Dataset for Category-Level Human-Object Interaction}},
author = {Yunze Liu and Yun Liu and Che Jiang and Kangbo Lyu and Weikang Wan and Hao Shen and Boqiang Liang and Zhoujie Fu and He Wang and Li Yi},
year = {2022},
eprint = {2203.01577},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2203.01577},
}
Affordance labels for ~1,800 household objects plus 50K intent-oriented hand-object interactions recorded on 100 of them; a common source for dexterous grasp synthesis. Follow-up OakInk2 covers bimanual tasks.为约 1800 个日常物体标注可供性，并在其中 100 个物体上记录 5 万次带意图的手-物交互；常用于灵巧抓取生成。后续 OakInk2 扩展到双手任务。
@misc{yang2022oakink,
title = {{OakInk: A Large-scale Knowledge Repository for Understanding Hand-Object Interaction}},
author = {Lixin Yang and Kailin Li and Xinyu Zhan and Fei Wu and Anran Xu and Liu Liu and Cewu Lu},
year = {2022},
eprint = {2203.15709},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2203.15709},
}
2.1M video frames of two hands manipulating articulated objects (scissors, laptops…) with accurate 3D hand and object meshes from motion capture; a reference for bimanual hand-object reconstruction.210 万帧双手操作铰接物体（剪刀、笔记本电脑等）的视频，借助动作捕捉提供精确的手与物体三维网格；双手-物体重建的常用基准。
@misc{fan2022arctic,
title = {{ARCTIC: A Dataset for Dexterous Bimanual Hand-Object Manipulation}},
author = {Zicong Fan and Omid Taheri and Dimitrios Tzionas and Muhammed Kocabas and Manuel Kaufmann and Michael J. Black and Otmar Hilliges},
year = {2022},
eprint = {2204.13662},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2204.13662},
}
Gym-style benchmark with cart-pole and quadrotor tasks, constraints and disturbances for comparing safe control and safe RL methods.Gym 风格基准，含倒立摆与四旋翼任务、约束与扰动，用于比较安全控制与安全强化学习方法。
@misc{yuan2021safe,
title = {{safe-control-gym: a Unified Benchmark Suite for Safe Learning-based Control and Reinforcement Learning in Robotics}},
author = {Zhaocong Yuan and Adam W. Hall and Siqi Zhou and Lukas Brunke and Melissa Greeff and Jacopo Panerati and Angela P. Schoellig},
year = {2021},
eprint = {2109.06325},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2109.06325},
}
@misc{grauman2021ego,
title = {{Ego4D: Around the World in 3,000 Hours of Egocentric Video}},
author = {Kristen Grauman and Andrew Westbury and Eugene Byrne and Zachary Chavis and Antonino Furnari and Rohit Girdhar and Jackson Hamburger and Hao Jiang and Miao Liu and Xingyu Liu and Miguel Martin and Tushar Nagarajan and Ilija Radosavovic and Santhosh Kumar Ramakrishnan and Fiona Ryan and Jayant Sharma and Michael Wray and Mengmeng Xu and Eric Zhongcong Xu and Chen Zhao and Siddhant Bansal and Dhruv Batra and Vincent Cartillier and Sean Crane and Tien Do and Morrie Doulaty and Akshay Erapalli and Christoph Feichtenhofer and Adriano Fragomeni and Qichen Fu and Abrham Gebreselasie and Cristina Gonzalez and James Hillis and Xuhua Huang and Yifei Huang and Wenqi Jia and Weslie Khoo and Jachym Kolar and Satwik Kottur and Anurag Kumar and Federico Landini and Chao Li and Yanghao Li and Zhenqiang Li and Karttikeya Mangalam and Raghava Modhugu and Jonathan Munro and Tullie Murrell and Takumi Nishiyasu and Will Price and others},
year = {2021},
eprint = {2110.07058},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2110.07058},
}
1,000 3D scans of real indoor spaces for Habitat; the main scene set behind ObjectNav and many embodied navigation benchmarks.1000 个真实室内空间的 3D 扫描，供 Habitat 使用；ObjectNav 等众多具身导航基准的主要场景集。
@misc{ramakrishnan2021habitat,
title = {{Habitat-Matterport 3D Dataset (HM3D): 1000 Large-scale 3D Environments for Embodied AI}},
author = {Santhosh K. Ramakrishnan and Aaron Gokaslan and Erik Wijmans and Oleksandr Maksymets and Alex Clegg and John Turner and Eric Undersander and Wojciech Galuba and Andrew Westbury and Angel X. Chang and Manolis Savva and Yili Zhao and Dhruv Batra},
year = {2021},
eprint = {2109.08238},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2109.08238},
}
@misc{chao2021dexycb,
title = {{DexYCB: A Benchmark for Capturing Hand Grasping of Objects}},
author = {Yu-Wei Chao and Wei Yang and Yu Xiang and Pavlo Molchanov and Ankur Handa and Jonathan Tremblay and Yashraj S. Narang and Karl Van Wyk and Umar Iqbal and Stan Birchfield and Jan Kautz and Dieter Fox},
year = {2021},
eprint = {2104.04631},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2104.04631},
}
VLN-CE moves Room-to-Room instructions from a fixed navigation graph into continuous Habitat scenes with low-level actions; the standard setting for realistic VLN.VLN-CE 把 Room-to-Room 指令从固定导航图搬到 Habitat 连续环境中、使用底层动作执行，是更贴近真实的 VLN 标准设定。
@misc{krantz2020beyond,
title = {{Beyond the Nav-Graph: Vision-and-Language Navigation in Continuous Environments}},
author = {Jacob Krantz and Erik Wijmans and Arjun Majumdar and Dhruv Batra and Stefan Lee},
year = {2020},
eprint = {2004.02857},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2004.02857},
}
EPIC-KITCHENS-100 — 100 hours (20M frames, ~90K action segments) of unscripted first-person kitchen video from 45 environments, with dense action labels and challenges such as action anticipation.EPIC-KITCHENS-100：来自 45 个厨房环境、100 小时（2000 万帧、约 9 万个动作片段）的非脚本第一视角视频，动作标注密集，并设有动作预测等挑战任务。
@misc{damen2020rescaling,
title = {{Rescaling Egocentric Vision}},
author = {Dima Damen and Hazel Doughty and Giovanni Maria Farinella and Antonino Furnari and Evangelos Kazakos and Jian Ma and Davide Moltisanti and Jonathan Munro and Toby Perrett and Will Price and Michael Wray},
year = {2020},
eprint = {2006.13256},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2006.13256},
}
An early shared dataset of robot interaction videos from seven robot platforms, used to pretrain visual dynamics models that transfer to new robots - a forerunner of Open X-Embodiment.早期的共享机器人交互视频数据集，涵盖七种机器人平台，用于预训练可迁移到新机器人的视觉动力学模型，是 Open X-Embodiment 的先驱。
@misc{dasari2019robonet,
title = {{RoboNet: Large-Scale Multi-Robot Learning}},
author = {Sudeep Dasari and Frederik Ebert and Stephen Tian and Suraj Nair and Bernadette Bucher and Karl Schmeckpeper and Siddharth Singh and Sergey Levine and Chelsea Finn},
year = {2019},
eprint = {1910.11215},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/1910.11215},
}
@misc{anderson2017vision,
title = {{Vision-and-Language Navigation: Interpreting visually-grounded navigation instructions in real environments}},
author = {Peter Anderson and Qi Wu and Damien Teney and Jake Bruce and Mark Johnson and Niko Sünderhauf and Ian Reid and Stephen Gould and Anton van den Hengel},
year = {2017},
eprint = {1711.07280},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/1711.07280},
}
The YCB object set — 73 everyday objects with scans, textured meshes and simulation models, plus benchmarking protocols; widely used in grasping and pose-estimation work.YCB 物体集：73 个日常物体，附扫描数据、带纹理网格与仿真模型，以及基准测试规程；被抓取与位姿估计研究广泛使用。
@misc{calli2015benchmarking,
title = {{Benchmarking in Manipulation Research: The YCB Object and Model Set and Benchmarking Protocols}},
author = {Berk Calli and Aaron Walsman and Arjun Singh and Siddhartha Srinivasa and Pieter Abbeel and Aaron M. Dollar},
year = {2015},
eprint = {1502.03143},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/1502.03143},
}
Elias Gläßer, Gülhan Şikaroğlu, Pavlo Kupyn, Marc Pollefeys et al.
Abstract摘要
A challenging bottleneck in manipulation research is reproducibility and comparability of methods in real-world environments. Since manipulation is an interactive task, it cannot be evaluated on a static dataset of observations. Benchmarking in simulation is a common remedy, but cannot cover the whole complexity of real-world tests. This paper proposes a new paradigm of stochastic challenges. Like the well-known YCB objects, it is based on reproducible real-world scenes and therefore allows system-level innovation and comparability between different platforms. Due to their stochastic nature, …
@misc{glaer2026benchmarking,
title = {{Benchmarking Real-World Robotic Manipulation through Stochastic Challenges}},
author = {Elias Gläßer and Gülhan Şikaroğlu and Pavlo Kupyn and Marc Pollefeys and Zuria Bauer and Hermann Blum},
year = {2026},
eprint = {2611.12804},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.12804},
}
Jinghe Wang, Xinrui Cao, Duo Wu, Chenghao Gu et al.
Abstract摘要
Robotic manipulation has advanced rapidly in mapping visual observations and language instructions to actions. However, reliable manipulation in open-world environments requires robust spatial understanding: the ability to identify targets in complex layouts (spatial reasoning) and adapt to scenes that continuously evolve through interaction (spatial memory). Yet existing embodied benchmarks fall short in evaluating spatial understanding: they often introduce strong visual cues that bypass spatial reasoning, and focus on static scenes that offer little insight into an agent's capacity for …
@misc{wang2026geomind,
title = {{GeoMind: A Benchmark for Spatial Understanding in Robotic Manipulation}},
author = {Jinghe Wang and Xinrui Cao and Duo Wu and Chenghao Gu and Yong Zhong and Linjia Kang and Tianyi Xiong and Zhi Wang},
year = {2026},
eprint = {2611.12175},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.12175},
}
Jiaxuan Guo, Xingyu Ren, Youran Sun, Jingxin Yang et al.
Abstract摘要
Robot demonstration datasets combine demonstrations from teleoperators of uneven experience, so collection teams need to know how many demonstrations from one group match those from another. We measure this with a data multiple: the ratio of the numbers of demonstrations two groups need for their policies to reach the same success rate. On robomimic's simulated Square and Can tasks, we train behavior-cloning policies on demonstrations from the two teleoperators robomimic labels Better and the two it labels Worse, varying dataset size and the share of Worse demonstrations. On Square, within …
@misc{guo2026relative,
title = {{The Relative Value of Demonstration Data: Execution Time Limits and Operator Mixtures in Behavior Cloning}},
author = {Jiaxuan Guo and Xingyu Ren and Youran Sun and Jingxin Yang and Kejia Zhang and Haizhao Yang},
year = {2026},
eprint = {2611.11947},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.11947},
}
Connected physical systems fuse observations from robots, IoT sensors, and edge-AI modules into a shared account of the world, yet most evaluations score only the final answer or action and cannot tell whether a method attributed the evidence, updated its belief, handled contradictions, and justified its decision. We argue that shared reasoning in Connected Robotics should be evaluated as an evidence-to-belief trajectory, and introduce the Connected Robotics Evidence Graph (CREG) Benchmark to make that trajectory measurable. CREG comprises 350 deterministically generated scenarios across …
Jiaxin Guo, Wenzhen Dong, Taoyu Wu, Haomin Kuang et al.
Abstract摘要
Understanding 3D scene dynamics in minimally invasive surgery (MIS) is a fundamental step toward physically intelligent and autonomous surgical systems. However, existing public datasets offer limited scalability and diversity in real-world clinical scenarios. Moreover, annotations typically focus on either tissue or instruments alone, overlooking tool-tissue interactions. This gap limits applications such as augmented reality guidance and autonomous surgery driven by vision-language-action (VLA) models or world models, which demand large-scale, high-quality 3D supervision. In this work, we …
@misc{guo2026surgtap,
title = {{SurgTAP-3D: A Large-Scale Cross-Surgery Video Dataset with Contact-Aware 3D Motion Tracking}},
author = {Jiaxin Guo and Wenzhen Dong and Taoyu Wu and Haomin Kuang and Mengya Xu and Kejian Shi and Ziyi Wang and Tongfan Guan and others},
year = {2026},
eprint = {2611.11697},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2611.11697},
}
Qinhong Zhou, Riley Zong, Chunru Lin, Diyi Hu et al.
Abstract摘要
Recent robot manipulation benchmarks have enabled systematic evaluation of general-purpose robot policies across diverse objects, scenes, instructions, and embodiments. However, most existing benchmarks still assume that the robot is the only active agent and that task semantics are determined primarily by the instruction and object state. This assumption overlooks a common source of real-world variation: humans actively sharing the workspace. In such settings, humans may reveal an implicit goal, complete part of the task, temporarily change object availability, or constrain safe robot motion …
@misc{zhou2026harp,
title = {{HARP-Bench: Benchmarking Robot Manipulation Around Active Humans}},
author = {Qinhong Zhou and Riley Zong and Chunru Lin and Diyi Hu and Shivraj Singh Bhatti and Hongxin Zhang and Lin Qiu and Hanqing Zeng and others},
year = {2026},
eprint = {2611.11123},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.11123},
}
Generative world modeling from large-scale visual data offers a promising path toward scalable robot manipulation. This calls for an intermediate representation that can be supervised at scale and remains actionable for downstream control. We propose 3D full-object flow, an object-centric motion representation that assigns dense 3D trajectories to points across the complete surface of a target object, including its occluded regions. By extending visible-surface object flow to the full object geometry, this representation makes complete object motion explicit and applies uniformly to both …
@misc{chen2026objectworld,
title = {{ObjectWorld: Scaling Object-Centric World Models with A Large-Scale 3D Full-Object Flow Dataset}},
author = {Haonan Chen and Sicheng Isabella Lu and Haitian Li and Jiajin Wu and Hexiang Wei and Jiawei Zhang and Xin Fei and Yiwen Hou and others},
year = {2026},
eprint = {2611.11085},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.11085},
}
Mayur Deshmukh, Aaron Tung, Junyi Li, Nikash Bhardwaj et al.
Abstract摘要
Humanoid robots must coordinate locomotion and manipulation to navigate complex environments and perform everyday tasks. Learning such behaviours from human demonstrations requires accurate whole-body motion paired with egocentric vision, particularly for lower-body movements and foot placements that are frequently invisible from a first-person perspective. However, existing egocentric datasets often lack high-fidelity optical motion capture or systematic coverage of diverse loco-manipulation behaviours. To address this gap, we introduce RoboMoCap, a large-scale multimodal dataset combining …
@misc{deshmukh2026robomocap,
title = {{RoboMoCap: A Large-Scale Whole-Body Motion Capture Dataset with Egocentric Vision for Robot Learning}},
author = {Mayur Deshmukh and Aaron Tung and Junyi Li and Nikash Bhardwaj and Ademi Adeniji and Vincent Liu and Michael Cho and Jiafei Duan},
year = {2026},
eprint = {2611.10994},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.10994},
}
Shao Yuan Chew Chia, Rebecca H. Jiang, Abhinav Agarwal, Andreas Hofmann et al.
Abstract摘要
The diversity of a distilled diffusion policy's outputs matters beyond how it impacts the policy's success rate. In particular, downstream applications like test-time search and policy improvement require exploration. Thus, we want to distill multi-modal demonstrators into diffusion policies which reproduce their conditional action distributions. Standard data collection for behavior cloning applies a sliding window to demonstration episodes, which pairs unique observation histories with actions. These datasets can be fit by a policy that is either multi-modal, or unimodal with high Lipschitz …
@misc{chia2026explicitly,
title = {{Explicitly capturing demonstrator multi-modality for distillation into diffusion policies}},
author = {Shao Yuan Chew Chia and Rebecca H. Jiang and Abhinav Agarwal and Andreas Hofmann and Leslie Pack Kaelbling and Russ Tedrake},
year = {2026},
eprint = {2611.10949},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.10949},
}
Pre-training large parametric "Vision-Language-Action" (VLA) models on large demonstration datasets has recently yielded promising results for general-purpose manipulation. However, these models still struggle under even small domain shifts relative to the training data, such as changes in camera viewpoint, background, object identity, or robot morphology. Motivated by the success of Retrieval-Augmented Generation (RAG) in language models, we pursue retrieval to boost VLA performance. Specifically, we seek to leverage retrieval from domain-mismatched, "in-the-wild" demonstration datasets like …
Rajat Bhattacharjya, Changwoon Lee, Hyeonjong Ju, Sing-Yao Wu et al.
Abstract摘要
Mobile robotic sensing systems must adapt to operator requirements, limited resources, changing connectivity, and observation opportunities shaped by physical motion. Evaluating runtime strategies therefore requires assessing whether their decisions fulfill mission objectives, beyond isolated perception accuracy or execution efficiency. We present EmbodiedIntentBench, a benchmark for intent-driven runtime adaptation in mobile robotic missions. It represents mission intent through structured contracts for coverage, temporal evidence, reporting, urgency, and resources, and couples simulated …
@misc{bhattacharjya2026embodiedintentbench,
title = {{EmbodiedIntentBench: Benchmarking Intent-Driven Runtime Adaptation for Mobile Robotic Missions}},
author = {Rajat Bhattacharjya and Changwoon Lee and Hyeonjong Ju and Sing-Yao Wu and Hamidreza Alikhani and Zhuoyi Yang and Ian Harris and Nalini Venkatasubramanian and others},
year = {2026},
eprint = {2611.10498},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.10498},
}
This note describes a multi-view video dataset of a deformable object driven by a measured input, with the measured pose of every camera, and a first 3D result obtained on it with LaGSplat. LaGSplat learns from video a low-dimensional latent state that is both the generalized coordinate of a learned dissipative Lagrangian and the conditioning variable of a Gaussian Splatting decoder; fitting the decoder in 3D calls for synchronized multi-view footage, which the original paper only had from synthetic renderings. Here a robot arm moves a flexible strip by oscillating along a one-dimensional …
@misc{pottier2026four,
title = {{A four-view smartphone video dataset of a robot-held flexible strip, with camera poses and a 3D LaGSplat baseline}},
author = {Louen Pottier},
year = {2026},
eprint = {2611.09959},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2611.09959},
}
When should a robot look at the cup, and when at the portafilter? Spatial-reasoning benchmarks evaluate whether a model can describe a scene, but manipulation needs something more specific: the ability to acquire the particular evidence that makes the current command executable. A view that is informative for one task can be useless for another, even at the same workstation. We introduce Seeing What Matters, a benchmark for task-conditioned actionable spatial belief in simulated tabletop manipulation. Across three domains--coffee workstation, drawer/cabinet, and microwave/appliance--the …
Grasp-feasibility predictors report near-perfect accuracy on Task and Motion Planning benchmarks, but high accuracy on these benchmarks does not verify that a predictor uses the collision geometry that the motion planner checks. We propose a planner-verified counterfactual (CF) test: holding the grasp pose fixed, we insert a solid obstacle into the approach corridor, add its points to the model input, and let a planner re-derive the label, which flips from feasible to infeasible. Across eight predictors spanning point clouds, voxels, depth, and object tokens, CF-pair AUROC falls from high …
Mohamed Elmoghany, Kehan Li, Ruoqu Chen, Tarun Chiruvolu et al.
Abstract摘要
Creative tool use is a fundamental human capability and an important step toward general-purpose embodied intelligence. It requires agents to reason about object affordances, physical constraints, and manipulation strategies, especially when standard tools are unavailable. However, progress in this area is limited by the lack of real-world datasets that capture human tool-use behavior with sufficient scale, fidelity, and diversity. Existing benchmarks are often simulation-based, focus on simplified interactions, or lack detailed observations and annotations. We introduce ToolDex, a …
@misc{elmoghany2026tooldex,
title = {{ToolDex: An Ego-Exo Human Demonstration Dataset for Creative Physical Tool Use}},
author = {Mohamed Elmoghany and Kehan Li and Ruoqu Chen and Tarun Chiruvolu and Shutong Zhang and Ryan Wang and Stephen Tian and Yaru Niu and others},
year = {2026},
eprint = {2611.09036},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.09036},
}
Jingdong Zhang, Jionghao Wang, Zhengming Yu, Yancheng Cai et al.
Abstract摘要
Frontier language models are increasingly deployed as agents, and how well an agent performs depends not only on its model but also on its harness: the prompts, tools, memory, and control loop through which the model acts. When an agent is applied to a new problem, a natural question is whether it can use the feedback of the task to improve the way it works, and not merely its answer. Most existing benchmarks keep the agent's harness fixed and score the solution it produces, or target large-scale model training. We introduce Flex-RSI, a benchmark for the recursive self-improvement of agents …
@misc{zhang2026flex,
title = {{Flex-RSI: Benchmarking Recursive Self-Improvement of Frontier Agents}},
author = {Jingdong Zhang and Jionghao Wang and Zhengming Yu and Yancheng Cai and Wei Chow and Chris Choy and Zhiyang Dou and Hang Hua and others},
year = {2026},
eprint = {2611.08946},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2611.08946},
}
Ramy ElMallah, Marc Bishara, Hyegang Kim, Yuxi Zhang et al.
Abstract摘要
Recent progress in embodied robotics is bringing robots closer to deployment in homes, as vision-language models (VLMs) are demonstrating significant advances in reasoning, making VLM-driven robots promising home assistant agents. These agents will often be introduced into homes that already have cameras, Internet of Things (IoT) devices, and residents with wearables. To fulfill their potential, they need to integrate with existing smart infrastructure and reason through ambiguous scenarios: a smoke alarm may come from a pan about to ignite, a pan already on fire, a resident cooking or a …
@misc{elmallah2026homeloop,
title = {{HomeLoop: A Simulation Framework and Benchmark for Robot-IoT Coordination in Everyday Domestic Assistance and Safety}},
author = {Ramy ElMallah and Marc Bishara and Hyegang Kim and Yuxi Zhang and Chongyu Zhu and Chi-Guhn Lee},
year = {2026},
eprint = {2611.08914},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.08914},
}
Wenbo Huang, Jianxin Bi, Shunyu Yang, Ruiyang Si et al.
Abstract摘要
Self-improving embodied agents are increasingly being developed and compared on robot simulation benchmarks. Several recent works expose the benchmark's success criteria to the learning agent, a practice we call Criteria Exposure (CE). These criteria, originally written to verify policies trained by imitation learning, often demand more than the corresponding language instructions, so a score from such a verifier may not faithfully reflect whether the instructed task was completed, especially for robot-control programs written by an agent. As embodied coding agents report ever higher …
@misc{huang2026learning,
title = {{Learning the Task or Learning the Verifier? Rethinking Criteria Exposure in Embodied Self-Improvement}},
author = {Wenbo Huang and Jianxin Bi and Shunyu Yang and Ruiyang Si and Rui Ni and Qiang Wang and Shulong Jiang and Yintianrun Zhang and others},
year = {2026},
eprint = {2611.08318},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.08318},
}
Satvik Sharma, Hengyuan Hu, Mihika Kamat, Dorsa Sadigh et al.
Abstract摘要
Improvements in multi-modal capabilities have enabled agents to act directly as robot policies, a setting we refer to as agentic robot control (ARC). We introduce RoboARCade, a benchmark of long-horizon manipulation tasks across three embodiments designed to test in-context adaptation, memory, and reasoning for precise control. The tasks build on existing open-source benchmarks, which we make harder by widening the range of initial conditions. RoboARCade is easy to use and extend: all tasks run on a single simulation backend, MuJoCo, each task is a self-contained data pack, and agents connect …
Fei Lin, Tengchao Zhang, Ziyang Gong, Xiaotong Yu et al.
Abstract摘要
Self-Driving Chemical Laboratories (SDCLs) are moving chemical experimentation toward embodied automation, where machines must perceive laboratory scenes, reason about experimental states, and act under physical, procedural, and safety constraints. Although Multimodal Large Language Models (MLLMs) are increasingly considered as embodied brains for such systems, existing laboratory benchmarks mainly focus on safety, anomaly detection, or general scientific understanding, leaving their execution-oriented embodied capabilities underexplored. To fill this gap, we introduce BEAKER, a Benchmark for …
@misc{lin2026beaker,
title = {{BEAKER: An Expert-Curated Benchmark for Embodied Brains in Self-Driving Chemical Laboratories}},
author = {Fei Lin and Tengchao Zhang and Ziyang Gong and Xiaotong Yu and Bohan Zhang and Yifan Zhou and Qihao Yang and Ji Dai and others},
year = {2026},
eprint = {2611.06911},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2611.06911},
}
Shuangyu Xie, Kaiyuan Chen, Uksang Yoo, Xinyue Zhu et al.
Abstract摘要
For robots to operate reliably in commercial and industrial applications, they must persistently perform specific tasks with known variations in object shape and pose. We call this Variational Automation (VA) to differentiate these settings from fixed automation. Inspired by recent papers on Agentic Robotics, where pretrained LLM and VLM coding agents combine interpretable robot programming with the open-world adaptability of model free policies, we propose a general set of agentic robotics benchmark design criteria that includes accessibility, repeatability, diversity, relevance to agentic …
@misc{xie2026vabar,
title = {{VABAR: A Variational Automation Benchmark for Agentic Robotics}},
author = {Shuangyu Xie and Kaiyuan Chen and Uksang Yoo and Xinyue Zhu and Meadow Shen and Jaimyn Drake and Eugen Solowjow and Lionel Gulich and others},
year = {2026},
eprint = {2611.06881},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.06881},
}
Sihang Chen, Chen Ye, Wei Chen, Xiaohan Lei et al.
Abstract摘要
In the real world, mobile manipulation robots must perform complex object interactions and make multi-step decisions, often relying on task-relevant information that is no longer observable. However, existing benchmarks provide limited evaluation of such memory-dependent decision-making in mobile manipulation. To address this gap, we introduce RoboCasa-Mem, a new benchmark comprising nine mobile manipulation tasks that require agents to use interaction history at key decision points. Specifically, the tasks are grouped into two categories of memory: One is environment memory, which captures …
@misc{chen2026robocasa,
title = {{Robocasa-Mem: Benchmarking Environment and Process Memory in Long-Horizon Mobile Manipulation}},
author = {Sihang Chen and Chen Ye and Wei Chen and Xiaohan Lei and Zhipeng Tang and Sha Zhang and Yuan Zhang and Zixuan Chen and others},
year = {2026},
eprint = {2611.06475},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.06475},
}
Manipulating diverse objects with multi-fingered dexterous hands is challenging due to the high dimensionality and complex dynamics. Human-Object Interaction (HOI) datasets provide rich knowledge about task information and embodied interactions. Instead of solely imitating the human demonstrations, our method learns to holistically predict future hand-object states by leveraging these datasets. The predicted future states of the object can serve as a general-purpose reward term for reinforcement learning, reducing reliance on task-specific reward engineering and enhancing generalization …
Rohit Mohan, Maximilian Luz, Swarnava Chowdhury, Yao Lu et al.
Abstract摘要
Holistic perception for autonomous agents requires reliably segmenting and associating scene elements across time and camera views, including objects outside a fixed semantic taxonomy. Existing open-set panoptic benchmarks focus on individual images or views, while panoramic tracking benchmarks assume closed-set taxonomies. Consequently, these benchmarks do not evaluate whether unknown objects can be consistently segmented and tracked across synchronized multi-camera sequences. We introduce WayPOP, a panoramic open-set panoptic tracking benchmark built on real Waymo PVPS sequences augmented …
@misc{mohan2026waypop,
title = {{WayPOP: A Panoramic Open-Set Panoptic Tracking Benchmark}},
author = {Rohit Mohan and Maximilian Luz and Swarnava Chowdhury and Yao Lu and Florian Drews and Thomas N\"urnberg and Yakov Miron and Federico Tombari and others},
year = {2026},
eprint = {2611.05801},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.05801},
}
Changyeon Lee, Kelin Yu, Zheng Wei, Shuna Ni et al.
Abstract摘要
Olfaction provides critical information about airborne chemical cues, yet it remains largely underexplored in embodied AI, with limited prior machine olfaction work focusing primarily on semantic odor perception. We introduce OdorSpaces, a physics-based simulation platform and benchmark for visual-olfactory embodied navigation. OdorSpaces augments Habitat-Sim with 3D odor transport and dispersion simulated by Fire Dynamics Simulator (FDS), enabling agents to navigate photorealistic indoor environments using both visual and olfactory observations. The benchmark includes two representative …
@misc{lee2026odorspaces,
title = {{OdorSpaces: Visual-Olfactory Embodied Navigation in 3D Environments}},
author = {Changyeon Lee and Kelin Yu and Zheng Wei and Shuna Ni and Ruohan Gao},
year = {2026},
eprint = {2611.05738},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.05738},
}
Albert Wilcox, Frank Chang, Nhi Nguyen, Aishani Chakraborty et al.
Abstract摘要
Despite growing interest in foundation models for robotic manipulation, systematic evaluation of their capabilities remains elusive. Existing benchmarks are either narrow and saturated, offering little signal on SOTA methods, broadly scoped, providing diversity at the cost of controlled experimentation, or require training on large, messy open-source datasets. We introduce MESA, an evaluation framework designed for precise, reproducible measurement of language-conditioned policy generalization. MESA provides five evaluation suites that isolate in-distribution performance and four distinct …
@misc{wilcox2026mesa,
title = {{MESA: An Evaluation Framework for Compositional, Semantic, and Spatial Generalization in Robotics}},
author = {Albert Wilcox and Frank Chang and Nhi Nguyen and Aishani Chakraborty and Jeremy A. Collins and Vaibhav Saxena and Benjamin Joffe and Siddharth Karamcheti and others},
year = {2026},
eprint = {2611.05567},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.05567},
}
Shivin Dass, Kevin Nguyen, Rohan Yelandur, Junhong Xu et al.
Abstract摘要
Imitation learning policies in robotics are trained on datasets collected through costly real-world teleoperation. Often, the data-collection process is unguided, relying on the teleoperator's intuition, resulting in data that contributes little to overall policy performance. Recent work has shown that curating such unguided datasets can substantially improve policy performance, but these approaches operate by filtering existing datasets after collection and therefore do not prevent the acquisition of uninformative data. In this work, we propose GUILD, a novel Guided Data Collection (GDC) …
@misc{dass2026guild,
title = {{GUILD: Guided Iterative Data Collection for Imitation Learning}},
author = {Shivin Dass and Kevin Nguyen and Rohan Yelandur and Junhong Xu and Roberto Mart\'in-Mart\'in},
year = {2026},
eprint = {2611.05368},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.05368},
}
Reliable evaluation of visual 3D reconstruction and depth estimation in vegetation-rich outdoor environments is limited by the scarcity of geometric ground-truth data and the cost of complete scanner surveys. We present a LiDARbased reference-generation pipeline for evaluating the zero-shot performance of feed-forward visual-geometry models alongside conventional structure-from-motion (SfM) methods. The pipeline reuses camera images, LiDAR scans, and reference poses from existing multi-sensor localization datasets to generate per-view depth maps and visible reference point clouds. To address …
@misc{zhong2026visibility,
title = {{Visibility-Aware Reference Generation for Benchmarking Visual 3D Reconstruction in Vegetation-Rich Environments}},
author = {Zhipeng Zhong and Matthew Coombes and Cunjia Liu},
year = {2026},
eprint = {2611.05361},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.05361},
}
Po-Yi Wu, Djengo Cyun-Jyun Fang, Dian Cheng, Tsung-Wei Ke
Abstract摘要
Reinforcement learning (RL) is promising for adapting robot policies to unstructured real-world environments. However, standard RL pipelines rely on episodic training with frequent scene resets, which is impractical for real-world deployment due to the need for substantial human intervention. While prior work has studied non-episodic RL in relatively small policies trained from scratch, its behavior when fine-tuning modern large-scale pre-trained vision-language-action (VLA) models remains underexplored. In this work, we introduce CRONOS, a simulation benchmark for RL fine-tuning of VLAs …
Sierra Bonilla, Fengyi Jiang, Chinedu Nwoye, Jingpei Lu et al.
Abstract摘要
We introduce iMED, the first synchronized multi-endoscope dataset for robot-assisted minimally invasive surgery (RAMIS), designed to address a fundamental limitation in surgical vision benchmarking: the absence of independent held-out viewpoints. Existing surgical datasets capture single-trajectory sequences from a single endoscope, making it impossible to distinguish geometric generalization from photometric interpolation along a narrow forward-facing path. iMED provides 340 sequences ~170K synchronized timepoints with 4 views per timepoint) recorded simultaneously from two independent …
@misc{bonilla2026imed,
title = {{iMED: A Multi-Endoscope Dataset for Surgical 3D Perception}},
author = {Sierra Bonilla and Fengyi Jiang and Chinedu Nwoye and Jingpei Lu and Kailey Reardon and Humphrey W. Chow and Francisco Vasconcelos and Sophia Bano and others},
year = {2026},
eprint = {2611.04755},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.04755},
}
Simulation benchmarks are the standard way to compare robot policies, but each benchmark brings its own evaluator, observation layout, and action convention, and evaluation is slow: 500 RoboTwin Easy episodes with one simulator and one model on one GPU take 359.7 minutes. Profiling the official StarVLA evaluator shows that simulation occupies CPU cores while the GPU averages 14.3% utilization. RoboScope Kit wraps each model once in a Python class called a Policy, and each benchmark once in an adapter around the benchmark's own evaluator, so the benchmark's rules and scoring stay unchanged. …
@misc{wu2026roboscope,
title = {{RoboScope Kit: Unified Interfaces for Fast and Scalable Embodied-AI Evaluation}},
author = {Deng-Ke Wu and Yifan Jin and Yida Yang and Lei Qian and Zhaoye Zhou and Hongru Zhang and Kai Ye and Shuyuan Hu and others},
year = {2026},
eprint = {2611.03850},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.03850},
}
Kobi Kelemen, Tomasz Niedzia{\l}kowski, Marcin Paszczak, Victoriano Montesinos
Abstract摘要
Training robot policies for multi-task pick and place requires demonstrations on many objects, which handheld Universal Manipulation Interfaces (UMIs) can collect quickly and cheaply. However, existing UMIs depend on external base stations, use rigid fingers that give no signal about contact forces, or use parallel-jaw fingers that limit grip strength. We present a battery-powered handheld interface, built on YUBI, whose soft, angled fingers wrap around objects, deform visibly under load as a visual proxy for force, and match the grippers of the robot. With it, we collect a 99.7-hour dataset …
@misc{kelemen2026robot,
title = {{Robot-Matched Soft Fingers for UMI Data Collection: Lessons from 100 Hours of Pick and Place}},
author = {Kobi Kelemen and Tomasz Niedzia{\l}kowski and Marcin Paszczak and Victoriano Montesinos},
year = {2026},
eprint = {2611.03779},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.03779},
}
Yang Qu, Qiang Fu, Zonghao Ying, Jiayi Zhao et al.
Abstract摘要
Embodied agents are increasingly deployed in open-world environments, yet current robustness evaluations remain biased toward hazardous instructions and jailbreak vulnerabilities. A more fundamental issue is largely overlooked: agents are rarely tested under semantically sufficient but contextually minimal instructions (i.e., common in real-world human communication), where auxiliary contextual cues are absent. For the first, we identify a novel phenomenon termed Ambient Context Illusion, where agents fail to execute tasks correctly when non-essential contextual details are omitted, despite …
@misc{qu2026vaguebench,
title = {{VagueBench: Benchmarking the Robustness of Embodied Agents on Ambient Context Illusions}},
author = {Yang Qu and Qiang Fu and Zonghao Ying and Jiayi Zhao and Yaoyuan Zhang and Jinyang Guo and Jian Yang and Aishan Liu and others},
year = {2026},
eprint = {2611.03765},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2611.03765},
}
Autonomous mobility is essential for sustained lunar surface operations, as communication delays and line-of-sight occlusions limit continuous teleoperation. Although recent traversability estimation methods increasingly rely on deep learning, the domain gap between terrestrial analogue environments and real lunar terrain can limit their generalisation, and real lunar traversability labels are scarce, making fully supervised training difficult. To address these challenges, we build a weakly supervised framework trained on sparse pairwise annotations and study whether augmenting visual …
We release all 3,718 episodes from the ArmnetBench v0.1 evaluation (arXiv:2607.24481), which contains 600 teleoperated demonstrations and 3,118 rollouts of seven imitation-learning and vision-language-action policies on 12 tabletop tasks, recorded on low-cost SO-101 arms; 73% of the rollouts are failures. Every policy for a task was trained or fine-tuned on the same 50 demonstrations and rolled out on the same cell, so the corpus contains many policies' failures on shared tasks. An operator labelled every rollout successful, suboptimal or failure. Each episode has three camera views, joint …
@misc{selvaraj2026rollout,
title = {{Rollout then Recycle: 3,718 Labelled Real-Robot Episodes from Benchmarking 7 Policies on 12 Tasks}},
author = {Praveen Selvaraj and Lorenzo Uttini and Ville Kuosmanen},
year = {2026},
eprint = {2611.03168},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.03168},
}
KlaskTron is an open-hardware physical platform for two-player adversarial reinforcement learning, with a learned actuator model that supports zero-shot sim-to-real transfer of trained policies. Building on this platform, we contribute KlaskTron-IL: a multi-tier imitation learning dataset (80k human demonstration timesteps collected directly on the physical robot, 320k symmetry-augmented timesteps, and synthetic expert trajectories up to 5M steps generated by the zero-shot-validated PPO policy), and diagnostic baselines that expose three characteristic failure modes of standard IL methods …
Real-world driving requires context-aware decision-making. Existing end-to-end driving benchmarks primarily assess trajectory quality through safety, compliance, comfort, and driving progress, leaving context-dependent decision-making insufficiently characterized. To fill this gap, we present PlanContext, a benchmark for evaluating in-context driving decision-making. PlanContext constructs paired evaluations by varying high-level driving intentions while keeping the observed scene fixed, enabling a natural and controllable assessment of context-dependent decision-making. We further define six …
@misc{li2026plancontext,
title = {{PlanContext: Benchmarking In-Context Driving Decision-Making}},
author = {Ruikai Li and Naiyu Gao and Hao Shan and Yizhe Fan and Yanhao Wu and Han Jiang and Yilong Ren and Zhenning Li and others},
year = {2026},
eprint = {2611.03067},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.03067},
}
Shivani Subramanyan, Saksham Sharma, Samahith SA, Ahaan Kotian et al.
Abstract摘要
Autonomous robots are gaining traction in off-road environments, driven by their rising demand in agriculture, defense, and search-and-rescue operations. However, these environments are inherently unpredictable and navigation remains a formidable challenge given the complex environmental conditions including challenging terrain and varying illumination. Off-road autonomy requires robust perception across multiple spatial scales and imaging modalities, yet most sensing setups struggle to capture both far-field environmental context and near-field terrain detail simultaneously across various …
@misc{subramanyan2026ms,
title = {{MS-DOTT: A Multi-Modal Dataset for Off-Road Traversability Estimation}},
author = {Shivani Subramanyan and Saksham Sharma and Samahith SA and Ahaan Kotian and Suresh Sundaram},
year = {2026},
eprint = {2611.02992},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.02992},
}
Stefano Trepella, Andrea Ostuni, Mauro Martini, Irene Borgini et al.
Abstract摘要
Social navigation is essential for autonomous service robots operating in human environments, but progress remains constrained by the lack of extensive real-world benchmarks. This paper presents the Real Social Navigation Testbed (RESONATE), a reproducible benchmark for human-aware navigation based on a unified robotic platform, a shared ROS 2 software stack, precise trajectory ground truth, and structured social interaction scenarios. We evaluate eight controller configurations, combining classical and predictive local planners with different social modeling strategies, over eight …
@misc{trepella2026resonate,
title = {{RESONATE: a Real Social Navigation Testbed for Autonomous Service Robots}},
author = {Stefano Trepella and Andrea Ostuni and Mauro Martini and Irene Borgini and Marcello Chiaberge},
year = {2026},
eprint = {2611.02937},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.02937},
}
Haoyuan Wang, Shuai Peng, Yongkai Liao, Yuxiao Sun et al.
Abstract摘要
Robust multimodal perception for soft robotic grasping is challenged by modality-dependent noise, semantic ambiguity, and limited coordinated multisensory data. We introduce Soma, a dataset containing 100 objects, 1,000 continuous grasping trajectories, and 74,914 multimodal samples grouped by software acquisition cycle, spanning remote RGB, hand RGB, depth, tactile, IMU, and structured text observations. We also collect Soma-Success, an independent grasp-outcome dataset with 1,000 trials (998 retained after quality control), to evaluate frozen-encoder transfer for post-hoc grasp outcome …
@misc{wang2026somaalign,
title = {{SomaAlign: Reliability-Aware Multimodal Representation Learning for Soft Robotic Grasping}},
author = {Haoyuan Wang and Shuai Peng and Yongkai Liao and Yuxiao Sun and Wei Luo and Xinxing Chen and Jian Huang},
year = {2026},
eprint = {2611.02794},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.02794},
}