The clipped policy-gradient method behind most GPU-parallel sim-to-real locomotion and dexterity training.采用截断目标的策略梯度方法，是大多数 GPU 并行仿真训练（运动控制、灵巧操作）背后的算法。
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},
}
RECAP trains a VLA with RL via advantage conditioning on demos, on-robot rollouts and teleop corrections; reports it more than doubles throughput on hard tasks like laundry folding and espresso making.RECAP 通过优势条件化，用演示、机器人自主运行数据和遥操作纠正对 VLA 做强化学习；论文报告在叠衣服、做意式咖啡等难任务上吞吐量提升一倍以上、失败率约减半。
@misc{intelligence2025vla,
title = {{π*0.6: a VLA That Learns From Experience}},
author = {Physical Intelligence and Ali Amin and Raichelle Aniceto and Ashwin Balakrishna and Kevin Black and Ken Conley and Grace Connors and James Darpinian and Karan Dhabalia and Jared DiCarlo and Danny Driess and Michael Equi and Adnan Esmail and Yunhao Fang and Chelsea Finn and Catherine Glossop and Thomas Godden and Ivan Goryachev and Lachy Groom and Hunter Hancock and Karol Hausman and Gashon Hussein and Brian Ichter and Szymon Jakubczak and Rowan Jen and Tim Jones and Ben Katz and Liyiming Ke and Chandra Kuchi and Marinda Lamb and Devin LeBlanc and Sergey Levine and Adrian Li-Bell and Yao Lu and Vishnu Mano and Mohith Mothukuri and Suraj Nair and Karl Pertsch and Allen Z. Ren and Charvi Sharma and Lucy Xiaoyang Shi and Laura Smith and Jost Tobias Springenberg and Kyle Stachowicz and Will Stoeckle and Alex Swerdlow and James Tanner and Marcel Torne and Quan Vuong and Anna Walling and others},
year = {2025},
eprint = {2511.14759},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2511.14759},
}
HIL-SERL — real-world RL with human corrections, reported to reach near-perfect success on precise assembly and dynamic tasks after a few hours of training.HIL-SERL：结合人类纠正的真机强化学习，报告称在精密装配与动态任务上经数小时训练即可接近满分成功率。
@misc{luo2024precise,
title = {{Precise and Dexterous Robotic Manipulation via Human-in-the-Loop Reinforcement Learning}},
author = {Jianlan Luo and Charles Xu and Jeffrey Wu and Sergey Levine},
year = {2024},
eprint = {2410.21845},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2410.21845},
}
Post-trains a VLA (GigaBrain-0.5, 10k+ h of robot data) with RL conditioned on a video world model's predictions (RAMP); reports ~30% over the RECAP baseline on laundry folding, box packing and espresso making.在 GigaBrain-0.5（预训练于 1 万多小时机器人操作数据）基础上，用以视频世界模型预测为条件的强化学习（RAMP）做后训练；论文报告在叠衣服、装箱和制作浓缩咖啡等难任务上比 RECAP 基线提升约 30%。
@misc{team2026gigabrain,
title = {{GigaBrain-0.5M*: a VLA That Learns From World Model-Based Reinforcement Learning}},
author = {GigaBrain Team and Boyuan Wang and Bohan Li and Chaojun Ni and Guan Huang and Guosheng Zhao and Hao Li and Jie Li and Jindi Lv and Jingyu Liu and Lv Feng and Mingming Yu and Peng Li and Qiuping Deng and Tianze Liu and Xinyu Zhou and Xinze Chen and Xiaofeng Wang and Yang Wang and Yifan Li and Yifei Nie and Yilong Li and Yukun Zhou and Yun Ye and Zhichao Liu and Zheng Zhu},
year = {2026},
eprint = {2602.12099},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2602.12099},
}
VLA that reasons in discrete motion tokens (a chain of dynamics) shared across embodiments; predicted futures are scored against what actually happened, so action-free human videos give RL signal. Sim and real tests.VLA 用跨本体共享的离散运动 token 链（Chain-of-Dynamics）表达动作思考；预测每条链导致的未来并与真实结果比对来强化，使没有动作标注的人类视频也能提供监督。在仿真与真机上验证人到机器人的迁移。
@misc{kim2026thinkworld,
title = {{ThinkWorld: World Imagination Reinforces Chain-of-Dynamics Reasoning}},
author = {Chaehyun Kim and Chi-Pin Huang and Jisoo Kim and Min-Hung Chen and Yu-Chiang Frank Wang and Seungryong Kim and Fu-En Yang},
year = {2026},
eprint = {2610.37666},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.37666},
}
DISCO runs progressive-widening MCTS in GPU-parallel simulation, with a learned policy proposing continuous actions, then distills plans into a policy-value model used without search. Reports wins over BC, RL and search.DISCO 在 GPU 并行物理仿真中用渐进展宽 MCTS 搜索长程重排计划，由学习的策略提出连续动作、价值函数打分，并周期性把搜索结果蒸馏回策略-价值模型；部署时无需仿真搜索。报告在仿真与真机重排任务上优于行为克隆、强化学习与搜索基线（CoRL 2026 Oral）。
@misc{saha2026long,
title = {{Long-Horizon Multi-Object Rearrangement via Distilled Tree Search over Continuous Actions}},
author = {Kallol Saha and Yining Hong and Amber Li and Yejin Choi and Maxim Likhachev and David Held},
year = {2026},
eprint = {2610.25785},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.25785},
}
Rather than distilling per-motion experts or brute-forcing one RL policy, trains a humanoid tracking policy jointly with one differentiable world model of shared dynamics, fit from simulator logs (CoRL 2026 oral).SuperMimic 不再蒸馏逐动作专家或用大算力硬训单一 RL 策略，而是把人形机器人动作跟踪策略与一个从仿真日志拟合的、共享动力学的可微世界模型联合训练（CoRL 2026 口头报告）。
@misc{liao2026supermimic,
title = {{SuperMimic: Scaling Humanoid Whole-Body Control with One Differentiable World Model}},
author = {Qiayuan Liao and Yiyang Shao and Junfeng Long and Kevin Zakka and Roberto Horowitz and Pieter Abbeel and Koushil Sreenath},
year = {2026},
eprint = {2610.28480},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.28480},
}
Rewinds an action-conditioned world model to key decision points to try alternative actions, turning outcomes into action-level supervision; real data keeps correcting the model. Reports +10.7–11.9 pts on 3 real tasks.把动作条件世界模型当作反事实引擎：回到关键决策点尝试其他动作并对比结果，得到动作级监督而无需单独的价值模型；真机经验再通过 LoRA 持续修正世界模型。报告在 3 个真实任务上比最强基线高 10.7–11.9 个百分点。
@misc{luo2026rollback,
title = {{Rollback: Experience Backtracking in World Models for Policy Improvement}},
author = {Jiayi Luo and Yang Zhang and Liliang Chen and Nan Wang and Chen Gao and Boxiang Qiu and Zhuoxuan Li and Baorui Peng and Chuankang Li and Yuxiang Yan and Si Liu and Qingyun Sun and Shuicheng Yan and Maoqing Yao and Jianxin Li and Guanghui Ren},
year = {2026},
eprint = {2610.36068},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.36068},
}
Deployed robots pool their runs; failures retrieve matching successful segments from the fleet buffer to fine-tune on, with no new demos. Reports 93–99% on five real tasks; 96 robots lift phone packing 79% → 98%.部署中的机器人共享执行经验：用失败片段在机群经验库中检索相似的成功片段来微调策略，无需新增示教或人工纠正。论文报告五个真机任务成功率 93–99%，96 台机器人把手机装盒任务从 79% 提升到 98%。
@misc{yu2026learning,
title = {{Learning from Fleet Experience: Collective Experience Learning Across 96 Robots}},
author = {Chao Yu and Shu'ang Yu and Hao Lin and Hongzhi Zang and Yuxiang Yan and Chenxi Jiang and Zhen Guo and Chuankang Li and Yihao Chen and Xin Xu and Xu Fu and Jiakai Zhou and Liangzhi Shi and Hongyi Ren and Yiwei Liu and Jing Wu and Yuanqing Wang and Chen Wu and Bo Dai and Chunyu Yang and Hongliang Huang and Shuoyu Yue and Weiqi Yin and Yinuo Chen and Ruihua Chen and Chunyang Zhu and Han Xiao and Yanghao Li and Ruipu Wu and Zhikai Cai and Xiu Chen and Xu Tan and Si Liu and Yuchen Yan and Jialin Zhang and Wen Zeng and Yiming Zhou and Zixiao Huang and Xinlei Chen and Guanghui Ren and Quanlu Zhang and Boxun Li and Maoqing Yao and Yu Wang},
year = {2026},
eprint = {2610.25736},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.25736},
}
Continued pre-training study for game-playing agents: 120 runs over 0.8B–27B models, 50–1,600 h of data and multi-world mixes; reports diversity aids transfer to related worlds, little to distant ones.针对游戏中的具身智能体做持续预训练的对照研究：120 组配置覆盖 0.8B–27B 模型、50–1600 小时数据和多世界混合；报告数据多样性有助于迁移到相近世界，对差异大的世界帮助有限。
@misc{zhang2026scaling,
title = {{Scaling in Games: Continued Pre-Training For Embodied Agents In Diverse Virtual Worlds}},
author = {Kuan Zhang and Yukun Chen and Zhihao Yang and Yue Su and Xiangnan Wu and Zirong Chen and Run Luo and Jinkun Hou and Tao Tan and Yinhe Zheng and Yiming Li},
year = {2026},
eprint = {2610.28698},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2610.28698},
}
CoRL 2026 Spotlight. Diffusion sequence prediction with selective re-noising, so segments of the horizon can be refined again using context from the rest.CoRL 2026 Spotlight。在扩散序列预测中选择性重新加噪，让时间跨度内的片段可以借助其余片段的上下文再次修正。
SCOPE gives diffusion trajectory models calibrated uncertainty tubes without Monte Carlo sampling, by distilling score curvature; tested in crowd navigation and Franka manipulation.SCOPE 通过蒸馏得分函数曲率，让扩散轨迹模型无需蒙特卡洛采样即可给出校准的不确定性管道；在人群导航和 Franka 操作上测试。
@misc{xue2026control,
title = {{Control-Ready Uncertainty for Trajectory Diffusion}},
author = {Zhiwei Xue and Jia Yue Kam and Jinhang Qiu and Yifeng Cheng and Ege Gursoy and Jiaming Wang and Vincent Bonnet and Harold Soh},
year = {2026},
eprint = {2610.12431},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.12431},
}
Federated RL for heterogeneous clients that aggregates diffusion behavior priors with optimal transport rather than averaging policies, plus a low-variance DICE value baseline.面向异构客户端的联邦强化学习：不再平均策略，而是用最优传输聚合扩散行为先验，并加入低方差的 DICE 价值基线。
Learns task-centric latent actions from videos (in DINO feature space, filtered by language) so one policy can be pretrained across robots and human video; reports beating OpenVLA with far less pretraining compute.在 DINO 特征空间中、借助语言指令过滤，从视频里学习“任务相关”的隐动作，使策略可跨机器人与人类视频预训练；报告以少得多的预训练算力超过 OpenVLA。
@misc{bu2025univla,
title = {{UniVLA: Learning to Act Anywhere with Task-centric Latent Actions}},
author = {Qingwen Bu and Yanting Yang and Jisong Cai and Shenyuan Gao and Guanghui Ren and Maoqing Yao and Ping Luo and Hongyang Li},
year = {2025},
eprint = {2505.06111},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2505.06111},
}
Online RL for VLAs with GRPO and only binary success rewards on OpenVLA-OFT, plus exploration tweaks; reports 99.1% on LIBERO and 70.4% on RoboTwin 1.0 (vs 39.8% for the SFT baseline).在 OpenVLA-OFT 上用 GRPO 与仅"成功/失败"二值奖励做在线强化学习，并加入若干促进探索的改动；论文报告 LIBERO 成功率 99.1%、RoboTwin 1.0 为 70.4%（SFT 基线 39.8%）。
@misc{li2025simplevla,
title = {{SimpleVLA-RL: Scaling VLA Training via Reinforcement Learning}},
author = {Haozhan Li and Yuxin Zuo and Jiale Yu and Yuhao Zhang and Zhaohui Yang and Kaiyan Zhang and Xuekai Zhu and Yuchen Zhang and Tianxing Chen and Ganqu Cui and Dehui Wang and Dingxiang Luo and Yuchen Fan and Youbang Sun and Jia Zeng and Jiangmiao Pang and Shanghang Zhang and Yu Wang and Yao Mu and Bowen Zhou and Ning Ding},
year = {2025},
eprint = {2509.09674},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2509.09674},
}
Open RL training framework for VLAs (PPO/GRPO across GPU-parallel simulators); reports 98.11% on 130 LIBERO tasks and 97.66% on 25 ManiSkill tasks after RL — a common base for VLA+RL work.面向 VLA 的开源强化学习训练框架（PPO/GRPO，支持 GPU 并行仿真器）；论文报告经 RL 后在 130 个 LIBERO 任务上成功率 98.11%、25 个 ManiSkill 任务上 97.66%，是 VLA+RL 研究常用的底座。
@misc{zang2025rlinf,
title = {{RLinf-VLA: A Unified and Efficient Framework for Reinforcement Learning of Vision-Language-Action Models}},
author = {Hongzhi Zang and Mingjie Wei and Si Xu and Yongji Wu and Zhen Guo and Yuanqing Wang and Hao Lin and Peihong Wang and Liangzhi Shi and Yuqing Xie and Zhexuan Xu and Zhihao Liu and Kang Chen and Wenhao Tang and Quanlu Zhang and Weinan Zhang and Chao Yu and Yu Wang},
year = {2025},
eprint = {2510.06710},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2510.06710},
}
Unitree's official starting point for RL locomotion on Go2, H1 and G1 — train in Isaac Gym, check sim-to-sim in MuJoCo, then deploy to the real robot.宇树官方的强化学习运动控制入门仓库，支持 Go2、H1、G1：在 Isaac Gym 中训练，在 MuJoCo 中做 sim2sim 验证，再部署到真机。
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},
}
Recasts RL as sequence modeling — a GPT-style model predicts actions conditioned on the desired return; an early bridge between transformers and control.把强化学习改写成序列建模：GPT 式模型以期望回报为条件预测动作；是 Transformer 与控制结合的早期代表。
@misc{chen2021decision,
title = {{Decision Transformer: Reinforcement Learning via Sequence Modeling}},
author = {Lili Chen and Kevin Lu and Aravind Rajeswaran and Kimin Lee and Aditya Grover and Michael Laskin and Pieter Abbeel and Aravind Srinivas and Igor Mordatch},
year = {2021},
eprint = {2106.01345},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2106.01345},
}
JAX physics engine that runs thousands of environments in parallel on GPU/TPU together with the RL training loop, cutting locomotion training from hours to minutes.基于 JAX 的物理引擎，在 GPU/TPU 上与强化学习训练循环一起并行运行数千个环境，把运动控制训练从数小时缩短到数分钟。
@misc{freeman2021brax,
title = {{Brax: A Differentiable Physics Engine for Large Scale Rigid Body Simulation}},
author = {C. Daniel Freeman and Erik Frey and Anton Raichuk and Sertan Girgin and Igor Mordatch and Olivier Bachem},
year = {2021},
eprint = {2106.13281},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2106.13281},
}
DeepMind's Python stack on top of MuJoCo — bindings, PyMJCF and Composer for building models and tasks, the standard Control Suite benchmarks, plus locomotion and arm-and-bricks manipulation tasks.DeepMind 基于 MuJoCo 的 Python 工具栈：底层绑定、用于程序化建模与编写任务的 PyMJCF 和 Composer、作为标准基准的 Control Suite，以及运动控制和机械臂拼积木操作任务。
@misc{tassa2020dm,
title = {{dm_control: Software and Tasks for Continuous Control}},
author = {Yuval Tassa and Saran Tunyasuvunakool and Alistair Muldal and Yotam Doron and Piotr Trochim and Siqi Liu and Steven Bohez and Josh Merel and Tom Erez and Timothy Lillicrap and Nicolas Heess},
year = {2020},
eprint = {2006.12983},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2006.12983},
}
TD3 — fixes DDPG's value overestimation with twin critics, delayed policy updates and target smoothing; still a standard continuous-control baseline.TD3：通过双 Critic、延迟策略更新和目标策略平滑解决 DDPG 的价值高估问题，至今仍是连续控制的常用基线。
@misc{fujimoto2018addressing,
title = {{Addressing Function Approximation Error in Actor-Critic Methods}},
author = {Scott Fujimoto and Herke van Hoof and David Meger},
year = {2018},
eprint = {1802.09477},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/1802.09477},
}
Tests six flow-matching VLAs and reports that one-step Euler action generation largely matches multi-step sampling at scale — a useful check before adding special one-shot objectives to speed up control (CoRL 2026).在六个基于流匹配的 VLA 上实验，报告单步 Euler 生成动作在规模足够时与多步采样效果基本相当——在为提速引入专门的单步生成目标之前值得先看（CoRL 2026）。
@misc{sujit2026when,
title = {{When One Step Is Enough: An Empirical Study of Euler Integration in Flow-Matching Policies}},
author = {Shivakanth Sujit and Luca Nunziante and Rousslan Fernand Julien Dossa and Kai Arulkumaran},
year = {2026},
eprint = {2610.30735},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.30735},
}
Instead of one fixed set of VLA weights, generates context-dependent low-rank updates for both the vision-language and action modules; reports gains up to 16.3% on LIBERO/Meta-World and up to 14% over π0.5 on real tasks.不再让 VLA 在所有交互情境下共用固定参数，而是按当前情境为视觉语言模块和动作模块生成耦合的低秩参数更新；报告在 LIBERO/Meta-World 上最高提升 16.3%，真机上相对 π0.5 最高提升 14%。
@misc{liu2026morphact,
title = {{MorphAct: Robot Policies Morphing with the Worlds They Act In}},
author = {Ziyan Liu and Jiting Liu and Weihong Zhong and Wenda Yu and Yeqiu Chen and Jason Li and Xiaojiang Peng and Shuyuan Hu and Kai Wang and Bo Zhao},
year = {2026},
eprint = {2610.20019},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.20019},
}
Practical warning: fine-tuning VLA weights in bf16 without an fp32 master copy rounds away AdamW updates below 1/512 of a weight; finds this recipe in 9 of 12 open VLA codebases and 79%+ of updates lost on LIBERO.实用警示：VLA 权重以 bf16 微调且不保留 fp32 主副本时，小于权重 1/512 的 AdamW 更新会被舍入抹掉；报告 12 个开源 VLA 代码库中 9 个的默认或文档配方如此训练，LIBERO 实验中至少 79% 的更新被抹除。
Pretrains a 4B VLA on 3D-grounded latent actions mined from 30K hours of human, robot and sim video, then adapts to real robot actions. Reports 98.8% on LIBERO and 88.6% on RoboTwin 2.0.从 3 万小时人类、机器人与仿真视频中学习带 3D 约束的连续潜在动作（RynnLAM），用于预训练 4B 参数 VLA，再适配到具体机器人动作空间。报告 LIBERO 98.8%、RoboTwin 2.0 平均 88.6%。
@misc{zhao2026rynnvla,
title = {{RynnVLA-Latent: Scaling Robotic Manipulation Foundation Models with 3D-Grounded Latent Actions}},
author = {Haoyu Zhao and Xingyue Zhao and Zhian Su and Bohan Hou and Yinzhou Tang and Fengyi Wang and Kehan Li and Yuming Jiang and Siteng Huang and Xin Li and Zhenyu Zeng and Mingxiu Chen and Deli Zhao and Zhongyu Li},
year = {2026},
eprint = {2610.34356},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.34356},
}
Meta-agent that improves a robot's harness by exploring: keeps a lineage graph of harness versions and runs new rollouts to probe failures before editing. Reports more success with fewer rollouts than static reflection.通过主动探索改进机器人 harness 的元智能体：维护 harness 版本谱系图，先主动生成新回放探查失败与恢复行为再修改；报告在 LIBERO-Pro、RoboCasa、RoboTwin 上比静态反思用更少仿真回放取得更高成功率。
@misc{zhang2026vista,
title = {{VISTA: Active Exploration for Self-Improvement in Embodied Agents}},
author = {Zhang Zhang and Shengjun Zhang and Simin Huang and Min Chen and Chong Xia and Xin Fei and Yueqi Duan},
year = {2026},
eprint = {2610.35051},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.35051},
}
Latent action model that fuses synchronized camera views to resolve depth and occlusion. Reports 17% better 6D action decodability than single-view and +13.5% policy success (vs +6.2%); human-hand demos help on hardware.融合多路同步相机视角的潜在动作模型（LAM），用交叉注意力解决深度与遮挡问题；报告 6D 动作可解码性比单视角高 17%，下游策略成功率提升 13.5%（单视角为 6.2%），真机上用人手演示也比纯动作策略高 20%。
@misc{hausdorfer2026multi,
title = {{Multi-View Latent Action Pretraining for Cross-Embodiment Robot Learning from Videos}},
author = {Oliver Hausdörfer and Felix Minzenmay and Mani Majd and Maximilian P. Christof and Shiyuan Zhang and Angela P. Schoellig},
year = {2026},
eprint = {2610.27794},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.27794},
}
Assigns VLA bit widths from short closed-loop task-success tests, not calibration sensitivity: ~4-bit weights on LIBERO across four models. Honest caveat: allocations transfer to a real arm for X-VLA but fail for π0.5.用短时闭环任务成功率（而非校准数据敏感度）为 VLA 各组件分配权重/激活位宽，并考虑组件间相互影响；在四个 VLA 与 LIBERO 全部套件上约 4 比特权重仍保持较强表现。作者如实指出：仿真选出的方案在真机上对 X-VLA 无损，对 π0.5 却失败。
@misc{xu2026quantize,
title = {{Quantize by Doing: Task-Guided Mixed-Precision Quantization of Vision-Language-Action Models}},
author = {Jiuyi Xu and Xiao Hu and Meida Chen and Song Wang and Yang Ye and Yangming Shi},
year = {2026},
eprint = {2610.38661},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.38661},
}
Turns a small open VLM into a code-as-policy agent: SFT, then multi-turn RL with a physical critic. Reports Qwen3.5-4B going from 0% to 73.8% on held-out LIBERO; code, data and weights released.把小型开放 VLM 训练成“代码即策略”机器人智能体的开源流水线：先 SFT，再做多轮 RL，并用读取仿真特权状态的物理评论家为每次请求分配贡献。报告 Qwen3.5-4B 在留出的 LIBERO 上由 0% 升到 73.8%，并公开代码、数据和权重。
@misc{fan2026captain,
title = {{CaPtain: A Pipeline for Post-Training Code-as-Policy Robot Agents}},
author = {Yijia Fan and Xilin He and Yilei Jiang and Dongming Wu and Jiaxu Wang and Xiangyu Yue},
year = {2026},
eprint = {2610.40799},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.40799},
}
Reframes VLA pretraining as goal-conditioned RL: contrastive learning gives the model a sense of whether a language goal is reachable. Pretrained on 167B tokens; strongest on long-horizon tasks.将 VLA 预训练重构为目标条件强化学习：用对比学习让模型判断语言目标是否可达。基于 1670 亿 token 预训练，在长时序任务上优势最明显。
@misc{zhang2026prts,
title = {{PRTS: A Primitive Reasoning and Tasking System via Contrastive Representations}},
author = {Yang Zhang and Jiangyuan Zhao and Chenyou Fan and Fangzheng Yan and Tian Li and Haitong Tang and Sen Fu and Xuan'er Wu and Qizhen Weng and Weinan Zhang and Xiu Li and Chi Zhang and Chenjia Bai and Xuelong Li},
year = {2026},
eprint = {2604.27472},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2604.27472},
}
Constrained decoding for LLM planners: grounded 'meta-tokens' guarantee syntax and preconditions by construction and cut decoder memory from over 107 TB to under 2 GB.面向 LLM 规划的约束解码：用接地的“元 token”从结构上保证语法与前置条件，并将解码内存从 107 TB 以上降到 2 GB 以内。
@misc{yidouweng2026meta,
title = {{Meta-Ctrl: Guaranteed Plan Generation by Decoupling Syntactic and Semantic Constraints}},
author = {Gwen Yidou-Weng and Edward Sun and Tianyi Ma and Metin Alp Dogan and Benjie Wang and Allen Peng and Guy Van den Broeck and Yuchen Cui},
year = {2026},
eprint = {2608.22149},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2608.22149},
}
Latent actions with polar structure in hyperbolic space: radius encodes how much changes, direction encodes what kind. Reports gains over latent-action baselines and pretrained VLAs.在双曲空间中构造极坐标结构的潜动作：半径表示变化多大，方向表示变化类型。报告优于其他潜动作方法和预训练 VLA。
@misc{jeong2026polar,
title = {{PoLAR: Factorizing Extent and Mode in Latent Actions for Robot Policy Learning}},
author = {Youngjoon Jeong and Jihwan Yu and Minsoo Jo and Junha Chun and Taesup Kim},
year = {2026},
eprint = {2606.21139},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.21139},
}
Weights each action chunk by how much it shortens the path to success on a reachability graph built over demos, rollouts and interventions. Reports 91% on precise tasks after a few offline rounds; no reward model.在示教、rollout 与人工干预构成的可达图上，按每个动作块缩短到成功的步数来加权行为克隆。报告经几轮离线迭代在精细任务上达到 91%，无需奖励模型。
@misc{li2026graph,
title = {{Graph-Reweighted Behavior Cloning with Human Interventions for Precise Robotic Manipulation}},
author = {Xiaoman Li and Zhiqiang Yang and Jiayu Hu and Jipeng Kong and Yang Zhang and Jingwei Zhang and Fangzheng Yan and Chunhua Shen and Chenjia Bai},
year = {2026},
eprint = {2610.20504},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.20504},
}
CTR retimes the two arms in bimanual demos at training time while keeping task-required constraints, so the policy stops using the other arm's progress as a shortcut cue.CTR 在训练时对双臂示教中两只手臂的动作时序重新组合，同时保留任务必需的约束，避免策略把另一只手臂的进度当作捷径线索。
@misc{hu2026learn,
title = {{Learn the Task, Not the Schedule: Reducing Spurious Cross-Arm Coupling in Bimanual Imitation Learning}},
author = {Shuyuan Hu and Deng-Ke Wu and Yifan Jin and Qianhang Luo and Mingchen Ju and Kai Ye and Bo Zhao},
year = {2026},
eprint = {2610.26507},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.26507},
}
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},
}
Mobile dexterous grasping learned in sim: a whole-body controller does reaching and active vision, local RL experts the object, all distilled into one point-cloud policy; reports zero-shot real transfer.在仿真中学习移动灵巧抓取：全身控制器负责底盘-手臂协同到达与主动视觉，局部强化学习专家负责物体交互，二者蒸馏成一个基于点云的视觉策略；报告可零样本迁移到真实世界的数百个物体与场景。
@misc{yang2026cowboy,
title = {{COWBOY: A Scalable Sim-to-Real Framework for Learning Contextual Whole-Body Manipulation}},
author = {Jiahui Yang and Jason Jingzhou Liu and Ray Muxin Liu and Andrew E. Wang and Kenneth Shaw and Deepak Pathak and Ruslan Salakhutdinov},
year = {2026},
eprint = {2610.28948},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.28948},
}
ARPG pretrains an action encoder on human manipulation data, then trains a 130M generative policy in its space. Reports bimanual dexterous success up from 10.1% (π0.5) to 44.7% on DexJoCo.ARPG 把动作表征学习与动作生成解耦：先在大规模人类操作数据上预训练动作编码器，再在其表征空间训练 1.3 亿参数的生成策略。报告在 DexJoCo 双手灵巧任务上平均成功率从 π0.5 的 10.1% 提升到 44.7%。
@misc{xu2026scaling,
title = {{Scaling Dexterous Manipulation via Action Representation Pretraining}},
author = {Yanjinghao Xu and Jiexin Zhang and Huan Zhao and Jianjia Mo and Zhenghao Yin and Dingkang Liang and Xiang Bai and Han Ding},
year = {2026},
eprint = {2610.34453},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.34453},
}
Language-to-dexterous-manipulation as modulated tracking: a distilled generalist tracker plus an RL-trained modulator that keeps adjusting the reference, object latent and position target; zero-shot sim-to-real.把语言条件灵巧操作变成“可调制的跟踪”：蒸馏得到的通用跟踪器加 RL 训练的调制器，在执行中持续修正参考轨迹、物体隐表征与位置目标，抑制误差累积；零样本迁移到真机。
@misc{adalibieke2026adadextrack,
title = {{AdaDexTrack: Dynamic Modulation for Adaptive and Generalizable Dexterous Manipulation Tracking}},
author = {Jianibieke Adalibieke and Qianwei Han and Xueyi Liu and Yuzhe Qin and Li Yi},
year = {2026},
eprint = {2610.33852},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.33852},
}
Foresight without predicting future frames: a training-only head forecasts the next action chunk and backpropagates through the policy's latent. No inference cost; reports up to +14 pts on mm-precision real tasks.不预测未来图像也能获得“预见性”：仅在训练时用级联头预测下一个动作块，并把损失回传到在线策略的隐状态，推理零额外开销；报告在耳机装配、穿鞋带等毫米级真机任务上最高提升 14 个百分点。
@misc{yang2026adapolicy,
title = {{AdaPolicy: Action-Driven Anticipatory Robot Policies via Next-Chunk Prediction}},
author = {Fan Yang and Yuting Su and Chen Zhao and Fugui Fan and Peiguang Jing},
year = {2026},
eprint = {2610.34280},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.34280},
}
@misc{strong2026real,
title = {{Real2Sem2Real: 3D Semantic Augmentations for Robot Learning from a Single Human RGB Demonstration}},
author = {Matthew Strong and Ankush K. Dhawan and Julien Poffet and Giuse Pham and Jinho So and Monroe Kennedy},
year = {2026},
eprint = {2610.28880},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.28880},
}
Sim-only RL recipe for in-hand object pose reaching on a LEAP hand, zero-shot to hardware; reports 95.2% of 4,800 unseen goals reached and a tunable residual safety filter that cuts drops by 75%.在仿真中无需人类数据训练 LEAP 灵巧手的手内物体位姿到达策略，零样本迁移到真机；报告在 4800 个未见目标上 95.2% 到达，可调节的残差安全滤波器使掉落减少 75%。
@misc{chuang2026dexterous,
title = {{Dexterous In-Hand Object Pose Reaching with Learned Safety Filters}},
author = {Ian Chuang and Hiya Shah and Vamsi Chilamkur and Iman Soltani and Claire Tomlin},
year = {2026},
eprint = {2610.37652},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.37652},
}
Learns history-aware progress rewards for long-horizon manipulation so missing prerequisites are not rewarded, then refines policies from BC to online RL. Reports 31.8% on 18 held-out RLBench tasks (+9.2 pts).为长时程操作学习依赖执行历史的进度奖励，避免跳过前置步骤也得高分，再从行为克隆过渡到在线强化学习微调策略；报告在 18 个未参与预训练的 RLBench 任务上平均 31.8%（高出最强基线 9.2 个百分点），真机 57.5%。
@misc{chen2026robolongr,
title = {{RoboLongR: Causally Consistent Rewards Teach Robots to Progressively Master Long-Horizon Manipulation}},
author = {Zixuan Chen and Lin Cao and Fenggang Zhang and Haotong Zhang and Yangtao Chen and Jieqi Shi and Jing Huo and Yang Gao},
year = {2026},
eprint = {2610.33863},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.33863},
}
Builds pseudo-bimanual demos from single-arm data by sagittal-plane mirroring plus LLM-written coordination constraints. On AgileX ALOHA it reports 2.11× cups collected vs. a single-arm baseline.将单臂示范沿矢状面镜像（观测、本体感觉与动作），再由 LLM 生成时空协调约束，构造伪双臂示范；在 AgileX ALOHA（RoboTwin 仿真与真机）上报告收集杯子数达到单臂基线的 2.11 倍。
@misc{wu2026robostd,
title = {{RoboSTD: Zero-Shot Single-to-Dual Transfer via Sagittal Mirroring for Bimanual Learning}},
author = {Hongyu Wu and Chong Liu and Shaoxuan Xie and Shaqi Luo and Hanyu Feng and Shihan Wu and Guocai Yao},
year = {2026},
eprint = {2610.36685},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.36685},
}
SpatialDex pre-trains a VLM-based dexterous policy on scene geometry and hand motion reconstructed from 1M RGB-only human clips. Reports +38.3 pts over prior methods on 6 real bimanual tasks.SpatialDex 从 100 万段纯 RGB 人类视频中重建场景几何与手部运动，并以几何作为输入和预测目标预训练基于 VLM 的灵巧策略；报告在 9 个仿真任务上比现有方法高 16.5 分、6 个真实双臂灵巧任务上高 38.3 分。
@misc{he2026reconstruction,
title = {{Reconstruction-Augmented Visuomotor Pre-Training of 3D-Grounded Bimanual Dexterity from RGB-Only Human Videos}},
author = {Xingyi He and Zhiwen Qiu and Adhitya Polavaram and Yunhao Cao and Cory Fan and Julius Samwer and Saleh Kalantari and Kuan Fang},
year = {2026},
eprint = {2610.29716},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.29716},
}
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},
}
Airflow extension for Isaac Lab with experimentally identified models, plus a light blower beside the gripper. Reports zero-shot real push-and-blow at 80% over 20 trials and 2.6° heading error with air alone.为 Isaac Lab 增加基于实测模型的气流扩展，并在夹爪旁加装轻型鼓风机。报告零调试真机部署：推吹混合策略 20 次成功率 80%，纯气流转向平均误差 2.6°。
@misc{yang2026beyondcontact,
title = {{BeyondContact: A Reinforcement Learning Framework for Airflow-Based Non-Contact Manipulation}},
author = {Peiyu Yang and Mohamad S. Moudallal and Antonio Langella and Giulio Turrisi and Victor Barasuol and Claudio Semini},
year = {2026},
eprint = {2611.03059},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.03059},
}
Helps RL explore contact-rich tasks by resetting episodes with a contact-aware constrained state sampler, covering diverse contact modes; trains universal in-hand and whole-body policies.用感知接触的约束状态采样器重置回合，覆盖多样接触模式，帮助 RL 探索接触丰富任务；可训练通用的手内和全身操作策略。
@misc{toussaint2026combining,
title = {{Combining Constrained Sampling and Reinforcement Learning for Manipulation}},
author = {Marc Toussaint and Cornelius V. Braun and Armand Jordana and Sayantan Auddy and Eckart Cobo-Briesewitz and Denis Shcherba and Tilman Burghoff and Justin Carpentier},
year = {2026},
eprint = {2602.08557},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2602.08557},
}
Updates a latent estimate of hidden object properties during execution to condition an imitation policy; helps with center-of-mass changes, though it reports only about 30% success on unseen locations in sim.执行中在线更新物体隐藏物性的潜变量，并以此调节模仿策略；对质心变化有帮助，但报告仿真中未见过的质心位置成功率仅约 30%。
Distills sim RL rollouts into a frozen flow-matching action decoder, then learns only a visual interface to its latent from a few demos. Reports beating fine-tuned π0.5 and GR00T N1.7 with 5-10x fewer demos.把仿真中强化学习的经验蒸馏进冻结的流匹配动作解码器，之后只需用少量示教学习从图像到其潜变量的视觉接口。报告用少 5–10 倍的示教超过微调后的 π0.5 与 GR00T N1.7。
@misc{wang2026latent,
title = {{Latent RL Priors for Data-Efficient Dexterous Manipulation}},
author = {Yixiao Wang and Peiyi Chen and Mustafa Mukadam and Shreekant Gayaka and Masayoshi Tomizuka and Son Tran},
year = {2026},
eprint = {2611.08458},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.08458},
}
Improves a frozen one-step generative policy offline by learning a residual on its input noise, trained with advantage-weighted reconstruction instead of critic gradients, so the base behavior does not drift.离线改进冻结的一步生成策略：学习对输入噪声的残差修正，用优势加权的动作重建来训练而不是用评论家梯度，从而避免偏离原有行为。
@misc{tang2026awls,
title = {{AWLS: Advantage-Weighted Latent Steering for One-Step Generative Robotic Manipulation Policies}},
author = {Jiahong Tang and Jiaju He and Zhiyuan Yang and Ruyang Han and Haiyue Zhu},
year = {2026},
eprint = {2610.27736},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.27736},
}
Hierarchical real-world RL for functional dexterous tasks: a hand skill handles contact, an arm-hand policy coordinates and triggers it. Reports 98.3% average success on six tasks within one hour of online training.面向功能性灵巧操作的分层真机强化学习：手部技能负责精确接触，臂-手协调策略负责手臂运动与技能触发。报告在六个任务上一小时在线训练即达 98.3% 平均成功率。
@misc{huang2026hifun,
title = {{HiFun: a Hierarchical Framework for Efficient Functional Dexterous Manipulation Learning}},
author = {Linyi Huang and Guowei Huai and Weibin Liu and Shulong Jiang and Ping Tan and Weixuan Zhang and Hui Zhang and Jie Song},
year = {2026},
eprint = {2610.19518},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.19518},
}
Meta-learned policy and forward model share a belief latent; at deployment the latent is updated from action-outcome mismatches, so the policy adapts to unknown dynamics on the fly.元学习的策略与前向动力学模型共享信念潜变量；部署时根据动作结果的偏差更新该潜变量，使策略能即时适应未知动力学。
@misc{li2026scouting,
title = {{Scouting the Dynamics Gap: Test-Time Policy Adaptation via Action-Outcome Feedback}},
author = {Yishu Li and Liyuan Geng and Xinyi Mao and Amber Li and David Held},
year = {2026},
eprint = {2609.36107},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.36107},
}
One shared object-pose reward, varied training scenes: RL experts discover regrasping, in-hand adjustment and handover on their own, then distil into one policy. Reports 61 of 80 real trials with unseen objects.只用统一的物体位姿奖励并变换训练环境，强化学习专家自行学会重抓、手内调整与双手交接，再蒸馏成单一策略。报告在未见物体上真实实验 80 次成功 61 次。
@misc{yan2026simduoreal,
title = {{SimDuoReal: Object-Centric Bimanual Policy for Emergent Dexterity}},
author = {Qiyang Yan and Yulin Liu and Akash Sharma and Haozhi Qi and Guanya Shi and C. Karen Liu},
year = {2026},
eprint = {2611.10410},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.10410},
}
Plugs several pretrained tactile foundation models into Diffusion Policy and compares them in one setup, including fine-tuning and mid-training strategies, with over 5,000 real-world rollouts.把多种预训练触觉基础模型接入 Diffusion Policy，在统一设置下比较，并考察编码器微调与任务专用中训练策略，含 5000 多次真实机器人回合。
@misc{murooka2026evaluating,
title = {{Evaluating Tactile Foundation Models for End-to-End Imitation Learning in Contact-Rich Manipulation}},
author = {Masaki Murooka and Ryoichi Nakajo and Tomohiro Motoda and Koshi Makihara and Hanbit Oh and Keisuke Shirai and Yukiyasu Domae},
year = {2026},
eprint = {2611.00957},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.00957},
}
TacEx points curiosity at touch: it splits model uncertainty by sense so exploration seeks contact, not free-space motion. The collected data trains pick-and-place offline and post-trains VLAs.TacEx 让好奇心聚焦触觉：按感知模态分解模型不确定性，使探索去寻找接触而非空中乱动。采集的数据可离线训练抓放策略，也可用于 VLA 后训练。
@misc{iten2026tactile,
title = {{Tactile Curiosity Drives Robot Interaction}},
author = {Klemens Iten and Alexander Proshkin and Bhavya Sukhija and Stelian Coros and Andreas Krause and Pieter Abbeel and Carmelo Sferrazza},
year = {2026},
eprint = {2609.40134},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.40134},
}
Fine-tunes humanoid motion-tracking policies online on the real robot: a proprioceptive neural dynamics model learned from real data supplies gradients, no motion capture needed. Shown on Unitree G1.在真机上在线微调人形机器人动作跟踪策略：用真实数据训练的本体感知神经动力学模型提供可反传梯度，无需动捕系统。在 Unitree G1 上验证对多种扰动的快速适应。
@misc{chen2026humanoidndm,
title = {{HumanoidNDM: Closing the Training-Deployment Gap of Motion Tracking via Humanoid Neural Dynamics Model}},
author = {Chao Chen and Zhenrong Wang and Jilong Wang and Yusen Feng and Jiangran Lyu and Li Yi and He Wang},
year = {2026},
eprint = {2610.34126},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.34126},
}
Smooth analytic contact gradients only in the backward pass, hard-contact forward rollout unchanged: a quadruped trotting policy trains in under 2 min on a desktop and transfers to hardware without fine-tuning.只在反向传播中换用平滑的解析摩擦接触梯度，前向仍是硬接触仿真：四足小跑策略在台式机上不到 2 分钟训练完成，无需微调即可迁移到真机；在接触隐式轨迹优化中也能找到更丰富的接触解。
@misc{kim2026diffcontact,
title = {{DiffContact: Differentiable Contact Dynamics for Policy Learning and Contact-Implicit Trajectory Optimization}},
author = {Gijeong Kim and Seungbin Oh and Dongyun Kang and Hae-Won Park},
year = {2026},
eprint = {2610.35852},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.35852},
}
Keeps a humanoid behavior foundation model frozen and trains a small lower-body residual adapter for stairs, gaps and platforms, distilled to onboard depth; tested on four trackers and on a Unitree G1 with SONIC.冻结人形行为基础模型，仅训练一个作用于下肢目标的轻量残差适配器来应对台阶、沟壑和平台，并从特权地形信息蒸馏到机载深度；在四种跟踪器上验证，并在 Unitree G1（SONIC）上真机部署。
@misc{li2026graft,
title = {{GRAFT: Geometry-Aware Residual Adaptation for Foundation Trackers}},
author = {Liang Li and Guoyao Zhang and Xingyu Chen and Sikai Wu and Qianyang Wu and Yuhang Lin and Renjing Xu and Mingliang Zhou and Diyun Xiang},
year = {2026},
eprint = {2610.36507},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.36507},
}
A stochastic motion tokenizer gives a high-level generator a compact interface to a whole-body tracking policy; reports better G1 motion reconstruction than FSQ/VQ-VAE and real dance/fight/walk sequences.用随机运动分词器把行为标签或文本生成的动作压缩成离散表示，直接作为全身跟踪策略的参考；报告在 Unitree G1 动作重建上优于 FSQ 与 VQ-VAE，并在真机上执行舞蹈、格斗、行走及 35 秒的格斗转舞蹈序列。
When a quadruped manipulator keeps failing, a coding agent writes an RL problem to learn the missing whole-body skill, adds it to a library and retries; transfers to a real robot (18/20 rack retrieval).四足机械臂反复失败时，由编码智能体判断缺少哪种全身控制能力，自动编写强化学习问题（奖励、课程、终止条件）训练新控制器并加入技能库再重试；习得扶墙抬升等技能，仿真五项超出初始能力的任务成功率 98.4–100%，无需真机训练即迁移到实体机器人（货架取物 18/20）。
@misc{liu2026elevate,
title = {{ELEVATE: Agentic Controller Acquisition from Task Failure for Quadrupedal Manipulation}},
author = {Shiqi Liu and Radu Corcodel and Ding Zhao},
year = {2026},
eprint = {2610.38388},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.38388},
}
Dynamic routing makes policy-update divergence misestimated in mixture-of-experts policies, so MAPO optimizes router and experts in turn. Reports tracking 182 motions PPO with MoE fails on, against one the other way.动态路由会让混合专家策略的更新散度估计失准，MAPO 因此交替优化路由与专家。报告可跟踪 182 个 PPO+MoE 失败的动作，反过来只有 1 个。
@misc{lim2026mapo,
title = {{MAPO: Mixture of Experts with Alternating Policy Optimization for Universal Humanoid Motion Tracking}},
author = {Seonghyeon Lim and Hyeonwoo Lee and Mincheol Kim and Kimin Lee and Hyun Myung},
year = {2026},
eprint = {2611.02613},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.02613},
}
Uses sample-based MPC in simulation as an automatic expert, then trains off-policy RL on sparse rewards, avoiding reward shaping; deployed on an arm-equipped Spot and a G1 humanoid.在仿真中用采样 MPC 自动生成专家数据，再以稀疏奖励训练离策略 RL，免去奖励塑形；已部署到带臂 Spot 和 G1 人形机器人。
@misc{schuck2026learning,
title = {{Learning Loco-Manipulation From SMPC Demonstrations With Sparse Offline-to-Online RL}},
author = {Martin Schuck and Maks Sorokin and Simone Manni and Duy Ta and Angela P. Schoellig and Marco Hutter and Simon Le Cleac'H and Jan Brüdigam},
year = {2026},
eprint = {2608.12063},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2608.12063},
}
Train the video world model's autoencoder for temporal predictability, not just reconstruction, with a past-conditioned decoder. Reports up to 5× cheaper inference, 7× faster training and lower long-horizon / OOD error.视频世界模型的潜空间不应只为重建而训练：加入时间可预测性目标并配合以历史为条件的解码器。报告推理成本最多降低 5 倍、训练快 7 倍以上，长时域与分布外误差降低 15–24%，与真实策略评测更一致。
Why action-conditioned video models ignore actions: expert demos carry too little action influence. AIM actively collects data to raise it; 500 such episodes + 3k demos match a 300k-demo model on controllability.解释动作条件视频世界模型为何忽略动作：专家示教的“动作影响力”太低，模型容量再大也无法弥补。AIM 主动采集数据提高这一指标；报告 500 条主动采集数据加 3000 条示教即可在末端可控性上追平 30 万条专家数据。
@misc{bateman2026training,
title = {{Training Controllable World Models via Action Influence Maximization}},
author = {Samuel M. Bateman and Tenny Yin and Chongyi Zheng and Lei Huang and Brian Wang and Benjamin Eysenbach and Jaime Fernández Fisac and Dhruv Shah},
year = {2026},
eprint = {2610.27387},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.27387},
}
Cross-embodiment video world model conditioned on RynnLAM's 3D latent actions, pretrained on 23 datasets (~553M frames) of human, robot and sim video; its generated data improves LIBERO and real-robot policies.以 RynnLAM 的三维潜在动作为统一条件接口的跨本体视频世界模型，在 23 个数据集（约 5.5 亿帧）的人类、机器人和仿真视频上预训练；生成的数据提升了 LIBERO 与真机策略成功率。
@misc{zhao2026rynnworld,
title = {{RynnWorld-Latent: A Cross-Embodiment World Model Conditioned on Universal Latent Actions}},
author = {Haoyu Zhao and Yinzhou Tang and Zixuan Wang and Minghao Zhu and Dongchi Huang and Xingyue Zhao and Qin Zhao and Kehan Li and Siteng Huang and Xin Li and Zhenyu Zeng and Mingxiu Chen and Deli Zhao and Zhongyu Li},
year = {2026},
eprint = {2610.36791},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.36791},
}
Shows a fixed world model's uncertainty can break when the deployment policy changes; reweights conformal calibration by a learned state-action ratio. Reports 90% sets covering 55.0% vs 90.9% on real Franka data.指出部署策略一变，固定世界模型的不确定性校准就可能失效；用域分类器学到的状态-动作密度比做加权共形校准，并给出覆盖率保证（部分用 Lean 4 形式化验证）。报告在真实 Franka 轨迹上，普通 90% 共形集合仅覆盖 55.0%，加权后为 90.9%。
@misc{mittal2026calibrated,
title = {{Calibrated for Which Policy? Conformal Uncertainty for Fixed World Models Under Intervention}},
author = {Ayush Mittal and Dhruv Gupta},
year = {2026},
eprint = {2610.35877},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2610.35877},
}
Steers a video world model toward rare but plausible bad outcomes (e.g. task failures) by optimizing its initial noise with a VLM objective, to evaluate and improve policies. Driving and manipulation.通过优化视频世界模型的初始噪声（以 VLM 为目标），把想象引向罕见但合理的不利结果（如任务失败），用于评估和改进策略；覆盖自动驾驶与操作。
@misc{seo2026stressdream,
title = {{StressDream: Steering Video World Models for Robust Policy Evaluation and Improvement}},
author = {Junwon Seo and Sushant Veer and Ran Tian and Wenhao Ding and Apoorva Sharma and Karen Leung and Edward Schmerling and Marco Pavone and Andrea Bajcsy},
year = {2026},
eprint = {2606.00267},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2606.00267},
}
Probe study of which video world model latents carry action information: temporal video pretraining matters, pixel reconstruction quality does not. V-JEPA-style encoders give the best trade-off.探测哪些视频世界模型的潜变量含有动作信息：关键在于时序视频预训练，而非像素重建质量；V-JEPA 类编码器取得最佳权衡。
@misc{yeom2026what,
title = {{What Makes Video World Model Latents Action-Relevant: Prediction over Reconstruction}},
author = {Jewon Yeom and Hanseul Kim and Jeongjae Park and Sungmok Jung and Jaejin Lee and Taesup Kim},
year = {2026},
eprint = {2606.07687},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2606.07687},
}
Off-route VLN agents keep following their stale history ('behavioral inertia'). MOON adds a deviation judge that clears it and a GRPO-trained recovery policy (<1% params); reports +5.2 / +6.0 points on R2R-CE / RxR-CE.发现 VLN 智能体偏离路线后会被历史记忆“惯性”拖住。MOON 给冻结的导航策略加上偏航判断器（判断何时清空历史）和 GRPO 训练的恢复策略，参数增加不到 1%；报告在 R2R-CE / RxR-CE 上分别提升 5.2 / 6.0 个百分点。
@misc{zhou2026moon,
title = {{MOON: Learning When to Forget and How to Move On in Vision-and-Language Navigation}},
author = {Jingmeng Zhou and Seson Sun and Binyao Xu and Lingming Zhang and Hanshi Wang and Fudong Ge and Weiming Hu and Jin Gao and Zhipeng Zhang},
year = {2026},
eprint = {2610.33546},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.33546},
}
MEND plans UAV-UGV recharge teamwork with masked diffusion, enforcing operational rules at execution through a logic-based action mask; tested on real maps and physical robot teams.MEND 用掩码扩散规划无人机与地面车协同充电任务，在执行时通过逻辑动作掩码强制满足运行约束；在真实地图和实体机器人队伍上测试。
@misc{deng2026collaborative,
title = {{Collaborative Planning for UAV-UGV Teams under Operational Constraints via Masked Error-bounded Neuro-Symbolic Diffusion}},
author = {Zihao Deng and Qianhuang Li and Maggie Wigness and John Rogers and Donghyun Kim and Hao Zhang},
year = {2026},
eprint = {2610.18850},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.18850},
}
Makes deep unfolding work for large conic solvers via matrix-free implicit differentiation and a stable PSD-projection gradient; reports up to 50× speedups on SDP-based planning.通过无矩阵隐式微分和稳定的半正定投影梯度，让深度展开适用于大规模锥优化求解器；报告在基于 SDP 的规划上最高提速 50 倍。
Continual off-policy RL for VLAs — small actor/critic networks carry over between tasks, distilled on a few kept episodes to avoid forgetting; reports 31–68.5% less real-robot time on new assembly tasks.面向 VLA 的持续离策略强化学习：轻量 actor/critic 在任务间延续，用少量保留的旧任务轨迹做蒸馏以防遗忘。论文报告新的接触式装配任务所需真机训练时间减少 31–68.5%，旧任务几乎不遗忘。
@misc{kim2026accrue,
title = {{ACCRUE: Continual Off-Policy Reinforcement Learning for Robot Foundation Models}},
author = {Changyeon Kim and Kimin Lee and Yashraj Narang and Yijie Guo},
year = {2026},
eprint = {2610.18596},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.18596},
}
Post-trains an open 9B robot planner (SFT + multi-turn RL) across harnesses that call VLA skills or write control code; reports LIBERO-PRO success rising from 18.1% to 71.0% without external memory.像训练编程智能体一样对 9B 开源机器人规划器做后训练（SFT + 多轮强化学习），适配调用 VLA 技能和生成控制代码两类执行框架；报告在 LIBERO-PRO 上无外部记忆成功率从 18.1% 提升到 71.0%。
@misc{liu2026agenticroborl,
title = {{AgenticRoboRL: Bringing Coding-Agent Post-Training to Agentic Robotics}},
author = {Xinxin Liu and Bingxin Xu and Wenrui Bao and Ming Li and Wentao Bao and Chen Chen and Yuzhang Shang},
year = {2026},
eprint = {2610.28589},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.28589},
}
Survey of recursive self-improvement in robot learning — improving skills, then the process that makes them — and when gains carry forward; notes no physical study has yet isolated gains from recursion itself.把"递归自我改进"放到机器人学习语境下的综述：从提升技能到改进产生技能的过程，并总结增益能持续累积的条件；指出现有真机研究尚未单独验证"递归"本身带来的收益。附论文列表。
@misc{luo2026toward,
title = {{Toward Recursive Self-Improvement in Physical AI}},
author = {Meng Luo and Zechen Bai and Shanqing Xu and Yanlin Li and Kaixin Li and Mengchen Qian and Weifeng Lu and Ran Zhang and Yingtian Zou and Ziyang Luo and Wei Chen and Hongzhan Lin},
year = {2026},
eprint = {2610.29677},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.29677},
}
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},
}
Distills a pretrained VLA teacher into a policy without action pretraining via on-policy supervision plus teacher-run recoveries, triggered by a reward model (AutoDAgger); reports up to 4.1x over BC on 3 real tasks.把预训练具身基础模型（教师）蒸馏到未做动作预训练的策略：在学生访问的状态上做在线监督，并由奖励模型监控进度、自动触发教师接管采集纠错轨迹（AutoDAgger，无需人工看护）；报告在 3 个真机任务上比行为克隆最高提升 4.1 倍。
@misc{lu2026tilde,
title = {{TILDE: Teacher-in-the-Loop Distillation from Embodied Foundation Models}},
author = {Yuxiang Lu and Yantong Liu and Jinghua Hou and Xueli Sun and Ruihua Han and Zhe Liu and Hengshuang Zhao},
year = {2026},
eprint = {2610.25376},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.25376},
}
Co-trains a VLA on robot demos and noisy egocentric-video actions by weighting each source by reliability and distilling from clean future frames; reports steeper scaling than the baseline up to 30k hours of data.将机器人示教与动作标签噪声较大的第一视角视频联合训练 VLA：按数据源可靠性分配监督权重，并从干净的未来画面蒸馏表征；报告在最多 3 万小时预训练数据范围内扩展斜率高于基线。
@misc{zhang2026egoforesight,
title = {{EgoForesight: Effectively Scaling Policy Learning with Noisy Egocentric Actions}},
author = {Chubin Zhang and Changyuan Wang and Yirui Wang and Ao Li and Shilin Ma and Yue Su and Cai Zhou and Xiuwei Xu and Tianle Zhang and Yansong Tang},
year = {2026},
eprint = {2610.31845},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.31845},
}
TIDE updates a pretrained policy online from human interventions and its own rollouts, with no critic or reward model; reports gains on four real bimanual tasks, each session finishing in under an hour.TIDE 在部署过程中利用人类干预和策略自身的执行数据在线更新预训练策略，无需价值函数或奖励模型；报告在四个真实双臂任务上优于初始策略，每次在线后训练不到一小时完成。
@misc{gu2026training,
title = {{Training in Deployment: Human-in-the-loop Online Post-Training for Robot Manipulation}},
author = {Chun Gu and Xiaofei Wei and Yiming Chen and Bozhou Zhang and Juntong Zhang and Xiatian Zhu and Li Zhang},
year = {2026},
eprint = {2610.29264},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.29264},
}
Uses RL as a demo generator, not a controller: a latent-conditioned policy finds several distinct successful solutions, exported as an imitation dataset; reports 78% real-robot success vs. 39% for the best baseline.把强化学习当作示教数据生成器而非部署控制器：以隐变量为条件的策略为同一任务找到多种不同的成功解法，导出为模仿学习数据集；报告真机下游成功率从最强基线的 39% 提升到 78%。
@misc{park2026discodemo,
title = {{DiscoDemo: Discovering Efficient and Diverse Robot Demonstrations for Imitation Learning}},
author = {Minho Park and Kinam Kim and Donghu Kim and Byungkun Lee and Dongyoon Hwang and Yongjae Shin and Junha Hyung and Hojoon Lee and Jaegul Choo},
year = {2026},
eprint = {2610.32093},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.32093},
}
Replay-free lifelong imitation learning: hierarchical dual-routing attention retrieves experts by instruction, then refines by context, plus key regularization against drift. Reports beating replay baselines on LIBERO.无回放的终身模仿学习：分层双路由注意力先按指令检索专家、再按上下文细化，并用键级几何正则抑制表征漂移带来的路由干扰；报告在多个 LIBERO 任务序列上优于无回放和有回放基线。
Fine-tunes diffusion and flow policies with automatic preference labels: rollouts are ranked by progress measured against a few expert demos, no human labels or reward design. Reports beating SFT on long-horizon tasks.用自动生成的偏好标签微调扩散与流匹配策略：以少量专家示教为参照衡量自主回放的顺序进度并据此排序，无需人工标注或手工奖励；报告在长时程操作任务上优于监督微调基线。
@misc{luu2026progress,
title = {{Progress-Guided Preference Generation for Aligning Generative Control Policies}},
author = {Tung M. Luu and Tri Ton and Duc M. Nguyen and Xuan Thanh Nguyen and Nguyen Gia Binh and Quang Tan Nguyen and Vu Duong and Daniel Sonntag and Duy M. H. Nguyen and Vien Anh Ngo and Chang D. Yoo},
year = {2026},
eprint = {2610.25692},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2610.25692},
}
FLEX lets language corrections edit a robot's task features via add, generalize and mask (not add-only). Reports higher normalized reward than other editing methods in 2D nav, Franka tabletop and a real user study.FLEX 用自然语言纠正来迭代修改机器人的任务特征集合，支持添加、泛化、屏蔽三种操作（而非只能添加）；在二维导航、Franka 桌面操作和真实用户研究中报告比其他特征编辑方法获得更高的归一化奖励。
@misc{lee2026flexible,
title = {{Flexible Language-Guided Representation Editing for Adaptive Robot Behavior}},
author = {Audrey Lee and Nathaniel Dennler and Ilia Sucholutsky and Andreea Bobu},
year = {2026},
eprint = {2610.28893},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.28893},
}
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},
}
Survey separating ordinary learner updates from improvers that revise how a robot learns, with an L0–L3 rubric from task-local adaptation to verified improver gain, and proposed comparisons from shared learner starts.综述：区分普通的学习器更新与会修改学习过程本身的“改进器”，给出从任务内适应到已验证改进增益的 L0–L3 分级，并提出从相同起点比较的评测方法。
@misc{han2026recursive,
title = {{Recursive Self-Improvement in Physical AI: Mechanisms, Evidence, and Evaluation}},
author = {Xiaofeng Han and Kaize Luo and Zenghuang Fu and Zichong Zhu and Zhe Feng and Changwei Wang and Rongtao Xu and Weiliang Meng and Xiaopeng Zhang},
year = {2026},
eprint = {2610.38266},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.38266},
}
CoRL 2026 Oral. Makes flow matching policy gradients work for robot control, bypassing action likelihoods, so policies need not be Gaussian. Applied to legged locomotion, humanoid motion tracking and manipulation.CoRL 2026 Oral。让流匹配策略梯度可用于机器人控制：绕过动作似然计算，策略不再限于高斯分布。应用于足式运动、人形动作跟踪与操作。
@misc{yi2026flow,
title = {{Flow Policy Gradients for Robot Control}},
author = {Brent Yi and Hongsuk Choi and Himanshu Gaurav Singh and Xiaoyu Huang and Takara E. Truong and Carmelo Sferrazza and Yi Ma and Rocky Duan and Pieter Abbeel and Guanya Shi and Karen Liu and Angjoo Kanazawa},
year = {2026},
eprint = {2602.02481},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2602.02481},
}
CoRL 2026 Oral. Adapts a VLA to expose a compact RL token, an interface for sample-efficient online RL, so a pretrained policy gains precision and speed from a few hours of real-world practice.CoRL 2026 Oral。让 VLA 输出紧凑的 RL token 作为在线强化学习接口，预训练策略只需几小时真实练习即可提升精度与速度。
@misc{xu2026rl,
title = {{RL Token: Bootstrapping Online RL with Vision-Language-Action Models}},
author = {Charles Xu and Jost Tobias Springenberg and Michael Equi and Ali Amin and Adnan Esmail and Sergey Levine and Liyiming Ke},
year = {2026},
eprint = {2604.23073},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2604.23073},
}
CoRL 2026 Spotlight. Adapts frozen generative policies from human interventions in latent space, as a cheaper alternative to mass data collection or online RL on hardware.CoRL 2026 Spotlight。在隐空间中根据人类干预适配冻结的生成式策略，替代大规模采集或实机在线强化学习。
@misc{murray2026flowdagger,
title = {{FlowDAgger: Human-in-the-Loop Adaptation of Generative Robot Policies in Latent Space}},
author = {Michael Murray and Daphne Chen and Simran Bagaria and Dean Fortier and Tess Hellebrekers and Galen Mullins and Harshavardhan Gajarla and Oier Mees and Maya Cakmak and Andrey Kolobov},
year = {2026},
eprint = {2607.08877},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2607.08877},
}
Prompt-only task-progress scoring: the VLM first describes what happens and defines absolute success, then scores with an explicit rubric. This curbs hallucinated progress on failed runs, and the rewards can drive RL.只靠提示词的任务进度评估：先让 VLM 客观描述画面并定义绝对成功标准，再按明确评分表打分，减少对失败执行“幻觉式”的进度判断，其奖励可用于下游强化学习。
@misc{memmel2026d,
title = {{D\&A: Decoupling Description and Assessment Enables VLMs as Zero-Shot Robotic Reward Models}},
author = {Marius Memmel and Shawn Yin and Urvi Rutia and Angad Wadhwa and Eldon Li and Gavin Angeloff and Ankit Goyal and Jesse Zhang and Dieter Fox and Fabio Ramos and Anqi Li and Abhishek Gupta},
year = {2026},
eprint = {2611.03622},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.03622},
}
Simulation study varying robot and object mass: RL copes, but behavior cloning does better when the robot is lighter than the object, and more data narrows the gap without closing it. Lower-inertia demos help.在仿真中改变机器人与物体质量：强化学习影响不大，行为克隆则在机器人比物体轻时更可靠，增加数据只能缩小而无法消除差距。采集低惯量示教可部分缓解。
Tracks closed-loop pass@k, the chance that one of k rollouts succeeds. Checkpoints with equal pass@1 can post-train differently, so it proposes offline checkpoint selection that balances coverage and reliability.提出跟踪闭环 pass@k（k 次尝试至少成功一次的概率）。pass@1 相同的检查点在后训练中可能表现不同，因此给出兼顾覆盖度与可靠性的离线检查点选择方法。
@misc{sun2026beyond,
title = {{Beyond Validation Loss: Coverage and Reliability Dynamics in Robot Policy Pretraining}},
author = {Zhanyi Sun and Max Simchowitz and Shuran Song},
year = {2026},
eprint = {2611.01565},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2611.01565},
}
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},
}
Online RL on top of a pretrained policy, penalised by distance from that policy's support: free inside it, smoothly costly outside; reports average success rising from 33% to 95% across 19 benchmark tasks.在预训练策略之上做在线强化学习，按动作到其支撑集的距离施加惩罚：集内不罚，集外平滑增长；报告 19 个基准任务的平均成功率从 33% 升至 95%。
@misc{xu2026gsbr,
title = {{GSBR: Geometric Support Behavior Regularization for Online RL}},
author = {Charles Xu and Jacob Berg and Liyiming Ke and Sergey Levine},
year = {2026},
eprint = {2611.05295},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.05295},
}
Shows termination-based constrained RL can learn to avoid the goal when constraints tighten near it, and fixes this with a dense success critic; validated on spacecraft docking tasks.发现基于终止的约束强化学习在目标附近约束收紧时会学会回避目标，并用稠密成功信号的辅助价值网络解决；在航天器对接任务上验证。
A policy predicts each action and how long to hold it. Reports matching high-frequency control at under half the rate on an RC car and Unitree Go1, plus simulated VLA inference.策略同时预测动作及其保持时长。报告在遥控车、Unitree Go1 和仿真 VLA 上以不到一半的控制频率达到高频控制的性能。
@misc{sukhija2025tarc,
title = {{TARC: Time-Adaptive Robotic Control}},
author = {Arnav Sukhija and Lenart Treven and Jin Cheng and Florian Dörfler and Stelian Coros and Andreas Krause},
year = {2025},
eprint = {2510.23176},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2510.23176},
}
ARPG has an LLM design the RL pipeline before deployment: MDP first, then simulator code, with checks that state, action and reward agree; reports more successful policies than monolithic LLM baselines in Isaac Sim.ARPG 让大模型在部署前设计强化学习流程：先定 MDP，再写仿真代码，并校验状态、动作与奖励是否一致；报告在 Isaac Sim 中成功率高于单体式大模型基线。
@misc{jegal2026llms,
title = {{LLMs as MDP Designers: A Dependency-Aware Agentic Framework for Automated Robot Policy Generation}},
author = {Hong Je-Gal and Jinwoo Kim and Chanbin Yi and Hyun-Suk Lee},
year = {2026},
eprint = {2611.04974},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.04974},
}
Ranks checkpoints without rollouts: action deviations are calibrated against state-dependent thresholds from held-out demos, then combined in a running product. Reports rankings that align with real deployment.无需实机回放即可排序检查点：用保留示教得到的状态相关阈值校准动作偏差，再按累乘聚合。论文报告其排序与真实机器人部署成功率一致。
Jointly trains a latent action model, latent policy and action decoder, so plentiful action-free video plus a few labelled robot demos can train a policy. Reports beating prior latent-action baselines.联合训练潜在动作模型、潜在策略和动作解码器，用大量无动作标注的视频加少量带标注示教即可训练策略。报告在仿真和真实任务上优于已有潜在动作方法。
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},
}
Inverse RL from observation only (videos, no actions or rewards) made practical on a real robot via online world modeling. Reports pick-and-place learned from scratch to 82% in under 40 minutes.仅从观测（视频，无动作和奖励）做逆强化学习，借助在线世界模型在真机上实现。报告不到 40 分钟从零学会抓放，成功率 82%。
@misc{han2026online,
title = {{Online World Modeling Enables Real-World Inverse Reinforcement Learning from Observation}},
author = {Tyler Han and Bat Nemekhbold and Siyang Shen and Rohan Baijal and Richard Ebock and Harine Ravichandiran and Sanghun Jung and Kevin Huang and Byron Boots},
year = {2026},
eprint = {2602.24121},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2602.24121},
}
Composed diffusion policies often learn identical components. A drop-in score-decorrelation loss makes each factor specialize, with no loss in multitask performance.组合式扩散策略的各个分量常常学成相同的副本。即插即用的得分去相关损失让各分量各自专门化，且不损失多任务性能。
@misc{laprevotte2026learning,
title = {{Learning to Be Different: Component Specialization for Compositional Diffusion Policies}},
author = {Alec Laprevotte and Haonan Chen and Xiaoshen Han and Yilun Du},
year = {2026},
eprint = {2610.27194},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.27194},
}
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},
}
Learns how a known symmetry acts on pixel observations and actions from offline data, then uses it for frame-averaged behavior cloning. Reports average success 80.5% to 85.1% on four Robosuite tasks.从离线交互数据中学习已知对称群如何同时作用于像素观测和动作，再用于帧平均行为克隆。报告在四个 Robosuite 任务上平均成功率从 80.5% 升到 85.1%。
@misc{liu2026mirror,
title = {{MIRROR: Reflecting Hidden Symmetries from Offline Robot Interaction Data}},
author = {Shiqi Liu and Yihang Yao and Peide Huang and Mengdi Xu and Ding Zhao},
year = {2026},
eprint = {2610.17422},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.17422},
}
Success Guided Sampling focuses massively parallel RL on task configurations at the edge of the policy's ability. With up to a million envs it solves quadruped terrain and assembly tasks; assembly transfers zero-shot.成功引导采样把大规模并行强化学习集中在策略能力边界附近的任务配置上。用最多约一百万个并行环境解决四足多地形与装配任务，装配策略零样本迁移到真机。
@misc{zhang2026balanced,
title = {{A Balanced Data Diet: Addressing the Exploration Bottleneck in Mega-Scale RL for Robot Control}},
author = {Octi Zhang and Mateo Guaman Castro and Patrick Yin and Ignacio Dagnino and Abhishek Gupta and Rosario Scalise and Byron Boots},
year = {2026},
eprint = {2610.12465},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.12465},
}
PPO-style fine-tuning for diffusion and flow policies that builds a surrogate policy ratio from forward noising and the score/flow loss, avoiding likelihoods. Tested in sim and real.为扩散与流匹配策略做 PPO 式微调：用前向加噪与得分/流匹配损失构造替代的策略比率，绕开似然计算；在仿真与真机上验证。
@misc{li2026robofft,
title = {{RoboFFT: Finetuning generative robot policy via online reinforcement learning with forward process}},
author = {Yu Li and Shenghe Hu and Yuhan Wang and Yaoxiang Pu and Haotong Zhang and Yuanpei Chen and Yaodong Yang},
year = {2026},
eprint = {2609.32236},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.32236},
}
Training-free reward editing: compares successful and failed rollouts in hindsight to penalize trap states that foundation-model rewards rate highly. Reports at least 3x the base policy's performance.免训练的奖励修正：事后对比成功与失败轨迹，惩罚被基础模型奖励高估、最终却导致失败的“陷阱状态”。报告性能至少是基础策略的 3 倍。
@misc{niu2026hire,
title = {{HiRE: Hindsight Reward Editing for Policy Finetuning}},
author = {Haoyi Niu and Zhengtao Han and Yufeng Ji and Zhongyu Li and Koushil Sreenath},
year = {2026},
eprint = {2609.27068},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.27068},
}
Post-training recipe that turns an off-the-shelf VLM into a free-form language reasoner guiding a low-level policy: mid-train on expert reasoning traces, then rubric-based RL from offline action data.把现成 VLM 训练成用自由语言推理来指导底层策略的“机器人推理器”：先用专家推理轨迹做中训练，再用离线动作数据做基于评分规则的单步强化学习。
@misc{wu2026r,
title = {{$R^3$: Training Robots to Reason in Natural Language via Reinforcement Learning}},
author = {Lehong Wu and Yuxiao Qu and Zheyuan Hu and Ivan Zhang and Limin Wei and Zackory Erickson and Aviral Kumar},
year = {2026},
eprint = {2608.26053},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2608.26053},
}
Bounded latent memory for imitation policies on tasks where a cue disappears before the decision; it is keyed by path signatures of the robot's own trajectory and attaches via adapters. Real-robot branching tasks.为“线索在决策前已消失”的任务设计的模仿策略记忆：固定大小的潜在记忆以机器人自身轨迹的路径签名为键读写，通过适配器接入现有策略。在真机分支任务上验证。
Uses 3D object motion as a shared interface: object motion and robot actions are modelled jointly with masking, so one pretrained model serves as policy, dynamics model and few-shot adapter. Robot, human and sim data.以 3D 物体运动轨迹作为统一接口，用掩码生成目标联合建模物体运动与机器人动作，同一套预训练参数可作为策略、动力学模型或少样本适配器。混合机器人、人类视频与仿真数据预训练。
@misc{cao2026unified,
title = {{Unified Motion-Action Modeling for Heterogeneous Robot Learning}},
author = {Yunhao Cao and Shitong Liu and Chao Feng and Meryl Zhang and Xuanchen Lu and Andrew Owens and Kuan Fang},
year = {2026},
eprint = {2606.16917},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.16917},
}
Fleet-scale offline-to-online RL that keeps post-training a generalist VLA from autonomous rollouts and human interventions. On 16 dual-arm robots and eight tasks it reports 95% average success.车队规模的离线到在线强化学习：利用自主执行与人工干预数据持续后训练通用 VLA。在 16 台双臂机器人、八个任务上报告平均成功率 95%。
@misc{wang2026learning,
title = {{Learning While Deploying: Fleet-Scale Reinforcement Learning for Generalist Robot Policies}},
author = {Yi Wang and Xinchen Li and Pengwei Xie and Pu Yang and Buqing Nie and Yunuo Cai and Qinglin Zhang and Chendi Qu and Jeffrey Wu and Jianheng Song and Xinlin Ren and Jingshun Huang and Mingjie Pan and Siyuan Feng and Zhi Chen and Jianlan Luo},
year = {2026},
eprint = {2605.00416},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2605.00416},
}
Finds that a few replay samples, where new and old observations look alike but need different actions, prevent most forgetting. Dropping just 10% of them raises forgetting more than 4.5x on LIBERO.发现回放缓冲中少数“观测相似但所需动作冲突”的样本（记忆锚点）承担了防遗忘的主要作用：仅剔除其中 10% 就会让 LIBERO 上的遗忘增加 4.5 倍以上。
@misc{du2026memory,
title = {{Memory Anchors for Continual Robot Learning}},
author = {Maximilian Du and Zhanyi Sun and Chen Xu and Paarth Shah and Masha Itkina and Shuran Song},
year = {2026},
eprint = {2608.26545},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2608.26545},
}
Agentic intervention model for human-in-the-loop RL: detects stalled exploration from value trends and steers the policy back to high-value states. Reports 57% fewer human interventions.用于人在回路强化学习的智能干预模型：根据价值变化趋势发现无效探索，并自动把策略拉回高价值状态。报告人工干预减少 57%，成功率提升 8.6%。
@misc{deng2026uniintervene,
title = {{UniIntervene: Agentic Intervention for Efficient Real-World Reinforcement Learning}},
author = {Haoyuan Deng and Yitong Gao and Yudong Lin and Haichao Liu and Zhenyu Wu and Ziwei Wang},
year = {2026},
eprint = {2606.12372},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.12372},
}
Studies how value-function reliability drives offline-to-online RL for manipulation, with a history-conditioned value estimator. Reports 86% on millimeter-level chip insertion.系统研究价值函数可靠性如何影响操作任务的离线到在线强化学习，提出历史条件价值估计器。报告毫米级芯片插装成功率 86%。
@misc{xia2026robo,
title = {{Robo-ValueRL: Reliable Value Estimation for Offline-to-Online Reinforcement Learning}},
author = {Wenke Xia and Pei Ren and Wenbo Yu and Yizhuo Zhang and Jifan Li and Yixue Zhang and Yinuo Zhao and Qingyang Gao and Jianlong Fu and Jian Tang and Ji-Rong Wen and Zhengping Che and Di Hu},
year = {2026},
eprint = {2607.09866},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2607.09866},
}
Discovers what a given robot can actually learn: a GFlowNet builds symbolic task skeletons, rewarded by how well low-level policies learn them in sim, then the tasks are distilled into one language-conditioned policy.自动发现某台机器人“学得会”的任务：用 GFlowNet 生成符号化任务骨架，以仿真中底层策略的学习效果作为奖励，再把发现的任务蒸馏进一个语言条件策略。
Harness that lets coding agents improve real robot policies on their own: automatic reset and verification, parallel rollouts and agent-driven code changes. Reports 99% success on tasks like fastening a zip tie.让编程智能体自主改进真实机器人策略的框架：自动复位与结果验证、多机并行评测，并由智能体分析日志、修改训练代码。报告在扎扎带等灵巧任务上成功率 99%。
@misc{xiao2026enpire,
title = {{ENPIRE: Agentic Robot Policy Self-Improvement in the Real World}},
author = {Wenli Xiao and Jia Xie and Tonghe Zhang and Haotian Lin and Letian "Max" Fu and Haoru Xue and Jalen Lu and Yi Yang and Cunxi Dai and Zi Wang and Jimmy Wu and Guanzhi Wang and S. Shankar Sastry and Ken Goldberg and Linxi "Jim" Fan and Yuke Zhu and Guanya Shi},
year = {2026},
eprint = {2606.19980},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2606.19980},
}
Shows discrete diffusion policies do real-time chunking natively, since inpainting is just unmasking: no fine-tuning, ~0.7x compute. Reports 65% higher success than flow-matching RTC on a real hockey task.指出离散扩散策略天然支持实时分块异步执行（补全即解掩码），无需微调、计算量约为 0.7 倍。报告在真机冰球防守任务上成功率比流匹配 RTC 高 65%。
@misc{wang2026discretertc,
title = {{DiscreteRTC: Discrete Diffusion Policies are Natural Asynchronous Executors}},
author = {Pengcheng Wang and Kaiwen Hong and Chensheng Peng and Katherine Driggs-Campbell and Masayoshi Tomizuka and Chenfeng Xu and Chen Tang},
year = {2026},
eprint = {2604.25050},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2604.25050},
}
Retries that learn from failure: preference pairs from failed attempts steer the policy away from them, plus small action perturbations, and feeds recoveries back into training. Reports +17.6% sim, +11.7% real.会从失败中学习的重试机制：用失败尝试构造偏好数据，引导策略避开失败行为，并加入小幅动作扰动做局部探索，成功恢复的轨迹再回流训练。报告仿真提升 17.6%，真机提升 11.7%。
Converts egocentric human videos into robot data: a video with the human replaced by a robot plus a feasible robot action trajectory. Claims the first zero-shot dexterous visuomotor policy with no real-robot demos.把第一人称人类视频转为机器人数据：生成把人替换成机器人的视频，以及满足可行性约束的机器人动作轨迹。声称首次在无真机演示下零样本训练出灵巧视觉运动策略。
@misc{liu2026egoengine,
title = {{EgoEngine: From Egocentric Human Videos to High-Fidelity Dexterous Robot Demonstrations}},
author = {Yangcen Liu and Shuo Cheng and Xinchen Yin and Woo Chul Shin and Alfred Cueva and Yiran Yang and Zhenyang Chen and Chuye Zhang and Danfei Xu},
year = {2026},
eprint = {2606.12604},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.12604},
}
Self-supervised failure detection from nominal demos only, using object-centric spatio-temporal scene graphs; both of its scores beat image-based detectors, especially on semantic and interaction failures.仅用正常演示做自监督失败检测：把操作场景建模为以物体为中心的时空图，所提两种失败评分都优于基于图像的检测器，尤其擅长语义和交互层面的失败。
One-step flow policy that puts observation, timestep and action tokens in one sequence for a single transformer. Reports 86.4% average success (+7.5 points) at 3.0 ms latency, 43.9x faster than DP3.一步生成的流策略：把观测、时间步与动作 token 拼成一个序列，由同一 Transformer 联合建模。报告平均成功率 86.4%（提升 7.5 个百分点），推理延迟 3.0 ms，比 DP3 快 43.9 倍。
@misc{zhou2026unified,
title = {{Unified Condition-Action Modeling for Accurate One-Step Action Generation}},
author = {Xinyu Zhou and Zikun Cai and Kuangji Zuo and Gen Li and Boyu Ma and Yanshuo Lu and Yutong Song and Mingqi Yuan and Jiayu Chen and Jianfei Yang},
year = {2026},
eprint = {2608.16153},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2608.16153},
}
Learns recovery skills with RL and, from them, discovers new relational predicates for an abstract planner, turning local recoveries into failure avoidance. Reports >50% gains over baselines in four sim domains.用强化学习学习失败恢复技能，并据此发现新的关系谓词、完善抽象规划模型，把局部恢复转化为全局避免失败。报告在四个仿真领域比基线高 50% 以上，并迁移到真机。
@misc{li2026recover,
title = {{Recover, Discover, Plan: Learning Skills and Concepts from Robot Failures}},
author = {Bowen Li and Mayank Mishra and Y. Isabel Liu and Stone Tao and Nishanth Kumar and Alexander G. Gray and Ruwan Wickramarachchi and Jonathan Francis and Sebastian Scherer and Tom Silver},
year = {2026},
eprint = {2606.18328},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.18328},
}
Uses suboptimal demos only at high and low diffusion noise levels, where their coarse or local structure still helps. Covers noisy, sim, mismatched and mixed data; reports up to 33% over co-training on Open X-Embodiment.只在扩散过程的高噪声与低噪声阶段使用次优演示，此时其粗粒度或局部结构仍然有用。覆盖带噪、仿真、任务不匹配和混合数据；报告在 Open X-Embodiment 上比协同训练基线高最多 33%。
@misc{wei2026ambient,
title = {{Ambient Diffusion Policy: Imitation Learning from Suboptimal Data in Robotics}},
author = {Adam Wei and Nicholas Pfaff and Thomas Cohn and Arif Kerem Dayı and Constantinos Daskalakis and Giannis Daras and Russ Tedrake},
year = {2026},
eprint = {2606.12365},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.12365},
}
Self-supervised progress reward learned from time-warped demos (speed changes, reversals), used to filter and reweight behaviour-cloning data. Reports up to ~18x T-shirt folding throughput over vanilla BC.用时间扭曲（变速、倒放）增强的演示自监督学习帧级进度奖励，再用它过滤和加权行为克隆数据。报告 T 恤折叠吞吐量最高为普通 BC 的约 18 倍。
@misc{yu2026warp,
title = {{WARP-RM: A Warp-Augmented Relative Progress Reward Model for Data Curation}},
author = {Justin Yu and Andrew Goldberg and Kavish Kondap and Karim El-Refai and Ethan Ransing and Qianzhong Chen and Mac Schwager and Fred Shentu and Philipp Wu and Ken Goldberg},
year = {2026},
eprint = {2606.28320},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.28320},
}
Retargets human whole-body demos to a mobile manipulator offline with a closed-form shoulder-elbow-wrist solver and lazy base control; claims the first zero-shot whole-body mobile manipulation from them.离线把人类全身演示重定向到移动操作机器人：闭式肩-肘-腕几何求解器加“惰性”底盘控制，得到一致的动作；声称首次仅凭离线人类演示实现零样本全身移动操作。
@misc{chen2026warp,
title = {{WARP: Whole-Body Retargeting for Learning from Offline Human Demonstrations}},
author = {Zhenyang Chen and Chuizheng Kong and Chuye Zhang and Yuanshao Yang and Lawrence Y. Zhu and Shreyas Kousik and Danfei Xu},
year = {2026},
eprint = {2606.29940},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.29940},
}
Shared latent action space grounded by predicting the same future across data sources, so action-free and differently-labelled data can train one BC policy. Reports +48% average success on five tasks.通过让不同数据源对未来观测的预测保持一致来“落地”一个共享潜在动作空间，使无动作标签和动作空间不同的数据都能训练同一个行为克隆策略。报告五个任务平均成功率提升 48%。
@misc{wang2026imitation,
title = {{Imitation from Heterogeneous Demonstrations using Grounded Latent-Action World Models}},
author = {Tianyou Wang and Anson Lei and Joe Watson and Ingmar Posner},
year = {2026},
eprint = {2606.21672},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.21672},
}
Learns a short-horizon 'what should happen next' intent model from internet human videos and pairs it with embodiment-specific policies trained in parallel sim; the same intent model drives a hand and a gripper.从互联网人类视频学习预测“场景接下来该发生什么”的短时意图模型，再与在大规模并行仿真中训练的本体专用策略结合；同一意图模型可驱动灵巧手和平行夹爪。
Treats head-camera motion in egocentric human video as a viewpoint action and pretrains active perception plus manipulation from it. Reports matching state-of-the-art models pretrained on robot data.把第一人称人类视频中的头部相机运动视为“视角动作”，同时预训练主动感知与操作能力。报告效果与使用机器人数据预训练的最先进模型相当。
@misc{lin2026activemimic,
title = {{ActiveMimic: Egocentric Video Pretraining with Active Perception}},
author = {Xingyao Lin and Guojin Zhong and Tianyi Lu and Ziyi Ye and Yichen Zhu and Zuxuan Wu and Yu-Gang Jiang},
year = {2026},
eprint = {2606.06194},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.06194},
}
In-context imitation policy trained only on RGB synthetic data, with next-subgoal image prediction. Reports 79% average success on 16 unseen real tasks from a single test-time demo.完全用 RGB 合成数据训练的上下文模仿学习策略，并预测下一个子目标图像。报告仅给一条演示时，在 16 个未见真机任务上平均成功率 79%。
@misc{qian2026synthicl,
title = {{SynthICL: Scalable In-context Imitation Learning with Synthetic Data}},
author = {Cheng Qian and Ruomeng Fan and Yifei Ren and Yilong Wang and Edward Johns},
year = {2026},
eprint = {2606.08154},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.08154},
}
Annotators define their own preference axes in language (speed, safety, care) and compare trajectories per axis; a language-conditioned reward trains a steerable policy. Reports +38 points over sparse and binary rewards.让标注者用自然语言自定义偏好维度（速度、安全、细致等）并逐维比较轨迹，训练语言条件奖励模型和可引导的策略。报告比稀疏奖励与二元偏好方法高 38 个百分点。
DDPG-style RL for flow-matching policies without backprop through the ODE: the critic gradient at a one-step action estimate is distilled into the velocity field. Reports 92% on dual-arm AirPods assembly.无需沿 ODE 反向传播的流匹配策略 DDPG 式强化学习：在一步估计的动作上计算 critic 梯度，并蒸馏进速度场。报告双臂 AirPods 装配成功率 92%，炒蛋任务评分 86%。
@misc{shi2026flowdpg,
title = {{FlowDPG: Deterministic Policy Gradient on Flow Matching Policies for Real-World Manipulation}},
author = {Kexin Shi and Junyao Shi and Poorvi Hebbar and Zhuolun Zhao and Tarun Amarnath and Yifan Su and Shikhar Bahl and Deepak Pathak},
year = {2026},
eprint = {2606.22303},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.22303},
}
Human-in-the-loop RL whose latent world model also predicts when the operator would take over, steering the policy away from risky states. On a 16-DoF LEAP Hand: 96.7% grasp success, up to 84% less intervention.人在回路强化学习：潜在世界模型额外预测操作者何时会接管，并据此让策略远离高风险状态。在 16 自由度 LEAP 手上抓取成功率 96.7%，人工干预最多减少 84%。
@misc{yin2026how,
title = {{How to Learn from What a Human Would Avoid? Intervention-Aware World Models with Real-World RL for Dexterous Manipulation}},
author = {Jiaju Yin and Zhenhui Zhang and Lixin Xu and Heng Zhang and Jun Shao and Yating Feng and Arash Ajoudani and Renjing Xu},
year = {2026},
eprint = {2609.06009},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.06009},
}
RL-trained AI coach that steps back as the learner improves, modelled as a game between learner and coach. A 33-person FPV drone-racing study shows faster human skill gains than prior AI coaching.用强化学习训练的 AI 教练：随学习者进步逐步减少辅助，把教练与学习者建模为非合作动态博弈。33 人第一视角无人机竞速研究显示，人类技能提升快于已有 AI 教练方法。
@misc{wang2026ai,
title = {{AI Coaching for Accelerating Human Skill Development with Reinforcement Learning}},
author = {Wei Wang and Enlin Gu and Antonio Loquercio and Haimin Hu and Rahul Mangharam},
year = {2026},
eprint = {2606.25337},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.25337},
}
Agent harness that automates sim RL from environment setup to trained policy: staged specialist agents, executable gates, parallel trials and reusable knowledge. 16 tasks across 6 benchmarks, some moved to real robots.自动化仿真强化学习全流程的智能体框架：从环境搭建、奖励设计到调参训练，由分阶段的专职智能体、可执行检查点、并行试验与经验复用完成。覆盖 6 个基准 16 个任务，部分迁移到真机。
@misc{li2026harbor,
title = {{HARBOR: A Harness Framework for Agentic Robot Reinforcement Learning}},
author = {Zechu Li and Yufeng Jin and Xiaoyang Liu and Puze Liu and Vignesh Prasad and Carlo D'Eramo and Georgia Chalvatzaki},
year = {2026},
eprint = {2606.08610},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.08610},
}
Attention pooling over intermediate CNN features, weighted by the robot's current proprioception, to keep control-relevant local detail. Reports +36.2% and +41.2% over pooling and explicit-crop methods.以机器人当前本体状态为查询，对 CNN 中间特征做注意力池化，保留与控制相关的局部信息。报告比常规池化与显式裁剪方法分别高 36.2% 和 41.2%，仅需训练 5.8% 的编码器参数。
@misc{wang2026localized,
title = {{Localized Visual Feature Aggregation via Focus Pooling for Visuomotor Policies}},
author = {Ruiyu Wang and Zheyu Zhuang and Danica Kragic and Florian T. Pokorny},
year = {2026},
eprint = {2609.08408},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.08408},
}
~60M-parameter policy with event memory for tasks where the deciding cue comes long before the action, plus the real-robot Camo-Dataset. Reports decision success up from 22.5% to 80.8% and 87.1% on LIBERO-10.约 6000 万参数的视觉运动策略，带事件记忆，面向“决定性线索早于动作很久出现”的任务，并发布真机基准 Camo-Dataset。报告决策成功率从 22.5% 提升到 80.8%，LIBERO-10 达 87.1%。
@misc{guo2026chameleon,
title = {{Chameleon: Control-Indexed Prospective Memory for Visuomotor Manipulation}},
author = {Xinying Guo and Chenxi Jiang and Hyun Bin Kim and Yuhang Han and Ying Sun and Yang Xiao and Jianfei Yang},
year = {2026},
eprint = {2603.24576},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2603.24576},
}
Action-to-action flow-matching policy conditioned on object-centric 'what' features and image-plane 'where' cues, improving robustness to distractors and object shifts while keeping few-step inference.以物体为中心的动作到动作流匹配策略：把观测拆成语义“是什么”特征与图像平面“在哪里”线索作为条件，在保持少步推理的同时提升对干扰物和位置偏移的鲁棒性。
@misc{li2026object,
title = {{Object-Centric Conditioning for Visuomotor Flow Matching}},
author = {Jijie Li and Xu Yang and Junhong Zou and Chunhai Zhao and Chaoyang Zhao and Zhen Lei and Xiangyu Zhu},
year = {2026},
eprint = {2609.24155},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.24155},
}
Human-in-the-loop RL that treats each intervention as a preference (human action over policy action) and corrects credit on suboptimal segments. Reports +24.5% average success and 1.3x faster convergence.人在回路强化学习：把每次人工干预视为偏好（人类动作优于策略动作），对次优片段做信用重新分配。报告五个真机任务平均成功率提升 24.5%，收敛快 1.3 倍。
@misc{liu2026preference,
title = {{Preference-Calibrated Human-in-the-Loop Reinforcement Learning for Robotic Manipulation}},
author = {Zeyi Liu and Guangyao Liu and Yinuo Qu and Yuquan Xue and Bofang Jia and Chunhua Yang and Weihua Gui and Keke Huang and Ziwei Wang},
year = {2026},
eprint = {2606.03949},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.03949},
}
Latent action model constrained to compose additively over short horizons (identity, inverse, cycle), so latents capture motion rather than scene detail and give better supervision for policies.对潜在动作空间施加短时可加组合约束（恒等、逆、循环一致），使潜在动作只编码运动而非场景细节，并为下游策略学习提供更好的监督。
@misc{wei2026learning,
title = {{Learning Additively Compositional Latent Actions for Embodied AI}},
author = {Hangxing Wei and Xiaoyu Chen and Chuheng Zhang and Tim Pearce and Jianyu Chen and Alex Lamb and Li Zhao and Jiang Bian},
year = {2026},
eprint = {2604.03340},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2604.03340},
}
Studies whether single-robot teleop data can train decentralised coordinated box-pushing; tests with up to 40 robots show coordination is hard to learn just by watching other robots.研究能否只用单机遥操作数据训练去中心化的多机协同推箱；多达 40 台机器人的实验表明，仅靠被动观察其他机器人很难学会协调。
@misc{matsumoto2026learning,
title = {{Learning Coordinated Visuomotor Box-Pushing from Solo Demonstrations}},
author = {Ryusei Matsumoto and Mei Kinjo and Yoko Sasaki and Keisuke Okumura},
year = {2026},
eprint = {2610.05677},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.05677},
}
Uses a VLM only at training time to pick task-relevant current and past information, distilled into a 'workspace token' the policy reads at deployment. Beats conditioning on curated past frames.仅在训练时用 VLM 挑出完成任务所需的当前与历史信息，蒸馏成策略部署时读取的“工作区 token”。效果优于直接输入精选历史帧。
@misc{dashora2026workspace,
title = {{Workspace Models: Lightweight Robotic Memory via Saliency-Driven Supervision}},
author = {Nitish Dashora and Douglas Chen and Idan Shenfeld and John Marangola and Pulkit Agrawal and Max Simchowitz},
year = {2026},
eprint = {2609.20820},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.20820},
}
Video comparator trained once on general videos that turns one successful demo into a dense progress reward, paying only for confident forward progress. Best RL success on 9 sim and 4 real tasks.只在通用视频上训练一次的视觉比较器，可把一条成功演示变成稠密的进度奖励，并只奖励置信度高的前进。在 9 个仿真和 4 个真机任务的强化学习中成功率最高。
@misc{yang2026rarm,
title = {{RARM: Confidence-Gated Progress Reward Modeling for RL in Manipulation}},
author = {Pengzhi Yang and Xinyu Wang and Pengyu Jing and Kehan Wen and Yiduo Qu and Zhenhao Huang and Minghao Fu and Xin Liu and Yaheng Shen and Fan Shi},
year = {2026},
eprint = {2606.22027},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.22027},
}
When teaching data under-specifies some reward features, the robot targets preference queries at those features and explains what to look at. Up to 30.8% better than standard active preference learning with few queries.当示教数据对某些奖励特征监督不足时，机器人把偏好询问集中在这些特征上，并附带解释提示人该关注什么。少量询问下比标准主动偏好学习高最多 30.8%。
Shows fixed rules for deriving labels from human feedback often disagree with what people mean, and learns these implications instead. Better preference estimates, incl. a real pizza-making study.发现从人类反馈推导未选动作标签的固定规则常与人的真实意图不符，改为学习这种“隐含含义”。偏好估计更准，并在真实做披萨协作实验中验证。
@misc{zhang2026rethinking,
title = {{Rethinking the Implications of Human Feedback for Preference Learning in Human-Robot Collaboration}},
author = {Qiping Zhang and Kate Candon and Debasmita Ghose and Marynel Vázquez},
year = {2026},
eprint = {2609.13982},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.13982},
}
Uses an imitation policy's predicted action chunks as short task-space references for dense progress rewards in sparse-reward RL, plus guided pretraining. Better early sample efficiency and fewer safety violations.把模仿策略预测的动作块转成短时末端参考轨迹，为稀疏奖励强化学习提供稠密进度奖励，并用于引导预训练。提升早期样本效率、减少安全违规。
LARBRID fine-tunes a pre-trained diffusion/flow policy via its initial noise, with a noise/action shared representation that fixes 'noise aliasing'. Reports up to 2-3x gains on locomotion and manipulation tasks.LARBRID 通过调整初始噪声来强化学习微调预训练的扩散/流策略，并学习噪声空间与动作空间共享的表示来消除“噪声混叠”问题。在运动与操作任务上报告最高 2–3 倍的提升。
@misc{liang2026efficient,
title = {{Efficient Diffusion Policy Fine Tuning with Latent Noise Representation Bridging}},
author = {Yuanchu Liang and Yue Yang and An-Chi He and Hanna Kurniawati},
year = {2026},
eprint = {2611.07275},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.07275},
}
Contrastive RL struggles when contact changes the dynamics mode. Interaction-weighted Resampling samples around before/during/after contact, reporting a 19.8% average gain in sim, including robot air hockey.接触会改变动力学模式，使对比式强化学习难以处理操作任务。交互加权重采样在接触前、中、后阶段重点采样，仿真中平均提升 19.8%，任务包括机器人空气曲棍球。
@misc{shen2026learning,
title = {{Learning Object Manipulation from Scratch via Contrastive Interaction}},
author = {Tongle Shen and Caleb Chuck and Fan Feng and Biwei Huang},
year = {2026},
eprint = {2606.11525},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.11525},
}
Runs rough actions (from a human or a VLM) backwards through a flow policy to their noise, then forward again to land on a nearby good action mode. Distilling this reports up to 95-point success gains in under a minute.把粗糙动作（来自人或 VLM 的指引）反向送入流策略得到其噪声，再正向生成，落到附近的优质动作模态。将其蒸馏后，不到一分钟训练即可报告最高 95 个百分点的成功率提升。
@misc{tang2026improving,
title = {{Improving Robotic Generalist Policies via Flow Reversal Steering}},
author = {Andy Tang and William Chen and Andrew Wagenmaker and Chelsea Finn and Sergey Levine},
year = {2026},
eprint = {2606.13675},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.13675},
}
Robots that never trained together share value functions, so a robot finishing a task leaves the workspace in a state the next robot prefers; a KNN uncertainty filter blocks overconfident values. Up to +48.5 pts.从未共同训练的机器人通过共享价值函数来协作，前一个机器人会把工作空间留在后一个机器人偏好的状态；KNN 不确定性过滤避免分布外的过度自信。成功率最高提升 48.5 个百分点。
@misc{andreu2026leave,
title = {{Leave It Better Than Found: Value-Guided Cooperation in Sequential Tasks}},
author = {Mario Gomez Andreu and Michael Amir and Antonio Marino and Tianxu An and Marco Hutter and Amanda Prorok},
year = {2026},
eprint = {2611.08075},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.08075},
}
Adapts a frozen generalist policy at inference: two small proxies (one copies base behaviour, one is fine-tuned on the task) and their velocity difference steers each denoising step. Base weights are not needed.推理时适配冻结的通用策略：训练两个小代理策略（一个模仿基座行为，一个在目标任务上微调），以二者的速度场差值引导每一步去噪。无需访问基座参数。
@misc{ning2026proxy,
title = {{Proxy Policy Steering}},
author = {Chuanruo Ning and Tianrui Wang and Wei-Chiu Ma and Kuan Fang},
year = {2026},
eprint = {2609.09148},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.09148},
}
For assistive devices with costly user trials: a forward model predicts user outcomes from control policies and drives active queries to learn each user's reward, then picks the final policy without more trials.面向用户试验代价高的辅助设备：先训练前向模型从控制策略预测用户结果，用它主动选择查询来学习个体奖励，再在不增加试验的情况下选出最终策略。
@misc{park2026simultaneous,
title = {{Simultaneous Forward and Inverse Human-in-the-Loop Optimization}},
author = {Kyeongwon Park and Steven H. Collins},
year = {2026},
eprint = {2609.22630},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.22630},
}
ReTVL learns a mistake-sensitive value function from imperfect demos, using sparse retry keypoints as pairwise preferences on top of global progress, then reweights demo chunks for behavior cloning.ReTVL 从不完美示教中学习对失误敏感的价值函数：在全局进度校准之上，以稀疏标注的“重试”关键点构造局部成对偏好，再用该价值为行为克隆的示教片段加权。
@misc{qin2026beyond,
title = {{Beyond Monotonic Progress: Retry-Supervised Value Learning for Robot Imitation}},
author = {Xinyao Qin and Junjie Lu and Kaixin Wang and Chuheng Zhang and Sinjae Kang and Kimin Lee and Min Xu and Bin Liang and Jun Yang and Li Zhao},
year = {2026},
eprint = {2606.24633},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.24633},
}
SCAPE predicts per-scenario real-world policy performance from a few paired sim/real runs plus many sim runs, correcting sim bias and calibrating with conformal prediction. Tested on driving and quadrupeds.SCAPE 利用少量成对的仿真/真机数据和大量仿真推演，按场景预测策略在真实世界的表现：先校正仿真标签偏差，再用共形预测校准不确定性。在自动驾驶与四足速度跟踪上验证。
Behavioral INR represents a policy as a state-to-action function modulated by an episode latent (FiLM), so policy identity in unlabeled multi-policy data can be inferred without labels.Behavioral INR 把策略表示为由回合级潜变量（FiLM）调制的“状态→动作”函数，从而在无标签的多策略行为数据中无监督地推断策略身份。
Tests why action chunking helps BC: not temporal consistency or horizon reduction, but non-Markovian expressivity and less compounding error (also matched by delayed policies), plus an 'implicit ensembling' effect.检验动作分块为何有助于行为克隆：并非时间一致性或缩短时域，而是非马尔可夫表达力与更少的误差累积（延迟策略也能做到），另外还有一种“隐式集成”效应。
@misc{lazzati2026why,
title = {{Why Does Action Chunking Improve Behavioral Cloning Performance in Robotic Control?}},
author = {Filippo Lazzati and Kyle Stachowicz and William Chen and Alberto Maria Metelli and Andrew Wagenmaker and Sergey Levine},
year = {2026},
eprint = {2608.02547},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2608.02547},
}
Finds real-system failures on a small budget by correcting cheap proxy (e.g. simulator) failure signals with residual modeling and choosing diverse, well-supported scenarios. Up to 2x the failures of random sampling.在有限测试预算下寻找真实系统的失败：用残差建模校正廉价代理（如仿真器）的失败信号，并挑选多样且有数据支撑的场景。发现的失败数最多为随机采样的 2 倍。
@misc{parashar2026coverage,
title = {{Coverage Aware Active Evaluation for Failure Discovery with Paired Systems}},
author = {Anjali Parashar and Rachel Luo and Apoorva Sharma and Sushant Veer and Edward Schmerling and Carson Sobolewski and Mingxin Yu and Chuchu Fan and Marco Pavone},
year = {2026},
eprint = {2608.13719},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2608.13719},
}
RoboFQE scores visual manipulation policies from offline trajectories, no rollouts: fitted Q-evaluation on a pretrained vision encoder plus learned success labels. Keeps rankings where video world models drift.RoboFQE 无需推演、直接用离线轨迹评估视觉操作策略：在预训练视觉编码器上做拟合 Q 评估并学习终止成功标注。在视频世界模型会累积误差的长时任务上仍能保持策略排序。
@misc{deng2026robofqe,
title = {{RoboFQE: Rollout-Free Evaluation of Visual Robot Policies from Offline Trajectories}},
author = {Yunfu Deng and Sangwoo Shin and Brahma S. Pavse and Josiah P. Hanna},
year = {2026},
eprint = {2610.18699},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.18699},
}
AnyRank ranks robot policies with sequential tests: trials go to hard-to-separate pairs, a (partial) ranking holds at any stopping time, and imperfect priors (sim, world model) save trials safely.AnyRank 用序贯检验为多个机器人策略排序：把试验分给难以区分的策略，任何时刻停止都能给出有误差保证的（部分）排序，可借助不完美的先验（仿真、世界模型）节省试验而不破坏保证。
@misc{badithela2026anyrank,
title = {{AnyRank: Reliable Anytime Policy Ranking via Sequential Testing}},
author = {Apurva Badithela and David Snyder and Tenny Yin and Jade Sceats and George Pappas and Nikolai Matni and Masha Itkina and Haruki Nishimura and Anirudha Majumdar},
year = {2026},
eprint = {2610.14603},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.14603},
}
Sleeping Robots: learn a new skill while 'awake', then consolidate one shared policy offline during 'sleep' from frozen critics or actor snapshots, combining gradients by Nash bargaining. +64% success on Meta-World MT5.Sleeping Robots：“清醒”时学习新技能，“睡眠”时利用冻结的评估器或策略快照离线巩固单一共享策略，并以纳什议价合并梯度。在 Meta-World MT5 上平均成功率提升 64%。
LeaP replaces the standard Gaussian source of a flow/diffusion-bridge policy with a proprioception-conditioned Gaussian over action chunks from a small MLP. 81.6% average on 15 RoboTwin tasks.LeaP 用一个小型 MLP 预测、以本体感觉为条件的动作块高斯分布，替换流匹配/扩散桥策略的标准高斯起点。在 15 个 RoboTwin 任务上平均成功率 81.6%。
@misc{dai2026where,
title = {{Where Should Action Generation Begin? A Learnable Source Prior for Generative Robot Policies}},
author = {Meipo Dai and Qiyuan Zhuang and He-Yang Xu and Ying-Jie Shuai and Yijun Wang and Qi Dou and Xiu-Shen Wei},
year = {2026},
eprint = {2606.17408},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.17408},
}
Feynman-Kac correctors with a new guiding potential push diffusion-policy samples toward rare but promising modes; trajectory optimization repairs them and they are added back to training.用 Feynman-Kac 校正器配合新的引导势能，把扩散策略的采样推向少见但有前景的模态；再以基于采样的轨迹优化修复这些轨迹，并加回训练集重训策略。
@misc{yu2026guided,
title = {{Guided Discovery of New Behaviors using Diffusion Policies}},
author = {Dian Yu and Sebastian Sanokowski and Majid Khadiv},
year = {2026},
eprint = {2606.08743},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.08743},
}
Turns ambiguous failed rollouts into recovery training data: a privileged coach proposes a corrected high-level decision, kept only if re-executing from the same state improves progress.把含义不明的失败推演转化为恢复训练数据：由特权教练推测早期的高层决策错误并提出修正，只有从同一状态重新执行确实提升进度时才保留。
MoRE 'unclones' unwanted behavior modes (e.g. handing a knife blade-first) by distilling a temporary mode classifier into the policy weights, with a retain loss. +44 pts success across eight tasks, no inference cost.MoRE 通过把临时模式分类器的重定向信号蒸馏进策略权重，并配合保持损失，“去克隆”不想要的行为模式（如刀刃朝前递刀）。八个任务平均成功率提升 44 个百分点，推理零开销。
@misc{wang2026behavior,
title = {{Behavior Uncloning: Distilling Mode Redirection into Policy Weights without Inference-Time Steering}},
author = {Hao Wang and Jiuzhou Lei and Dayou Li and Bangya Liu and Minghui Zheng and Manling Li and Ruohan Zhang and Zhiwen Fan},
year = {2026},
eprint = {2606.29201},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.29201},
}
OTQL post-trains and speeds up a suboptimal flow policy from the robot's own experience with advantage-weighted conditional optimal-transport flow matching, in 50-60 episodes and without distillation.OTQL 用优势加权的条件最优传输流匹配，从机器人自身经验对次优流策略做强化学习后训练并加速推理，只需 50–60 个交互回合，无需蒸馏。
@misc{sochopoulos2026optimal,
title = {{Optimal Transport Q-Learning for Flow Policy Steering and Acceleration}},
author = {Andreas Sochopoulos and Esmeralda S. Whitammer and Nikolaos Tsagkas and João Moura and Michael Gienger and Sethu Vijayakumar},
year = {2026},
eprint = {2607.06262},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2607.06262},
}
ExpertGen trains a diffusion prior on imperfect demos (even LLM-synthesized), then RL steers its initial noise with the policy frozen, keeping behaviour human-like under sparse rewards; experts feed sim-to-real BC.ExpertGen 先在不完美示教（甚至由大模型合成）上训练扩散先验，再冻结策略、用强化学习只优化初始噪声，在稀疏奖励下保持类人行为；得到的专家策略用于仿真到真实的行为克隆。
@misc{xu2026expertgen,
title = {{ExpertGen: Scalable Sim-to-Real Expert Policy Learning from Imperfect Behavior Priors}},
author = {Zifan Xu and Ran Gong and Maria Vittoria Minniti and Kausik Sivakumar and Ahmet Salih Gundogdu and Eric Rosen and Riedana Yan and Tushar Kusnur and Zixing Wang and Di Deng and Peter Stone and Xiaohan Zhang and Karl Schmeckpeper},
year = {2026},
eprint = {2603.15956},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2603.15956},
}
A VLM agent explores simulated tasks by calling reusable RL-trained interaction skills, verifying outcomes and keeping memory in a verifier-guided tree search, to generate robot data without scripts.VLM 智能体在仿真中调用可复用的强化学习交互技能来探索任务，借助验证、反思与记忆进行验证器引导的树搜索，无需脚本即可生成机器人数据。
@misc{zhu2026skillweaver,
title = {{SkillWeaver: Agentic Exploration over Neural Interaction Skills for Scalable Robot Data Generation}},
author = {He Zhu and Lusen Zhao and Kwan Man Cheng and Su Li and Katerina Fragkiadaki},
year = {2026},
eprint = {2609.36171},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.36171},
}
GUILD decides what to collect next: a VLM proposes environments, object setups and goals, collected data is evaluated for its value to the policy, and the loop repeats, instead of filtering after the fact.GUILD 主动决定下一步采集什么数据：由 VLM 提出环境、物体布置与目标，采集后评估数据对策略的价值，再进入下一轮，而非事后筛选数据集。
@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 Martín-Martín},
year = {2026},
eprint = {2611.05368},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.05368},
}
SDPG estimates policy gradients from random perturbations of rollouts, so visual RL needs far fewer batch-rendered envs and trains in hours on one RTX 4080; adds visual benchmarks and sim-to-real demos.SDPG 通过对推演轨迹随机扰动来估计策略梯度，视觉强化学习所需的批量渲染环境大幅减少，单张 RTX 4080 几小时即可训练；并提供视觉机器人基准与仿真到真实演示。
@misc{you2026efficient,
title = {{Efficient On-policy Visual-RL via Stochastic Decoupled Policy Gradient}},
author = {Haoxiang You and Yilang Liu and Davis Zong and Qian Wang and Teeratham Vitchutripop and Qi Wang and Daniel Rakita and Ian Abraham},
year = {2026},
eprint = {2605.26478},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2605.26478},
}
Argues policies should condition on both the executed robot state and the leader command while predicting future commands, since their mismatch signals contact; beats either alone with no force sensing.认为策略应同时以机器人实际执行状态和主端指令为输入并预测未来指令，因为两者的偏差蕴含接触信息；在没有力觉输入的低成本平台上优于单独使用任一信号。
@misc{xu2026mind,
title = {{Mind the Gap: Rethinking I/O Design for Contact-Rich Visuomotor Policy Learning}},
author = {Cuijie Xu and Shurui Zheng and Zihao Su and Zhongchen Jian and Yuanfan Xu and Tinghao Yi and Xudong Zhang and Jian Wang and Yu Wang and Jinchen Yu},
year = {2026},
eprint = {2602.08776},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2602.08776},
}
Real users intervene when progress stalls, often with imperfect fixes; a progress-based human model makes sim benchmarking match them. Q-OIL adds optimistic bonuses on intervention steps; reports 10-50% higher success.真实用户往往在进展停滞时介入，且纠正动作并不完美；据此建立的基于进展的人类模型让仿真评测更贴近真人。Q-OIL 对介入转移加乐观奖励并用解耦评论家稳定训练，报告成功率提升 10-50%。
@misc{teegavarapu2026q,
title = {{Q-OIL: Optimistic Exploration in Learning from Human Interventions}},
author = {Pranav Teegavarapu and Sriyash Poddar and Rico Qi and Cleah Winston and Krisha Khandelwal and Aykut Onyol and Natasha Jaques and Abhishek Gupta},
year = {2026},
eprint = {2611.08576},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.08576},
}
Diffusion policies trained on sliding-window demos often memorise one action per state; querying the demonstrator k times at one state (vine sampling) keeps the multimodality and aids best-of-N and expert iteration.用滑窗示范训练的扩散策略往往每个状态只记住一个动作。在同一状态下向示范者查询 k 次（vine 采样）可保留其多模态分布，并提升 best-of-N 与专家迭代效果。
@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},
}
Post-trains a vision foundation model for robot control: a temporal-abstraction loss keeps action-relevant change, patch alignment to the original keeps semantics. Gains on several benchmarks and a real arm.不从头预训练，而是对现成视觉基础模型做面向机器人控制的后训练：时间抽象目标保留与动作相关的变化，与原模型的 patch 级对齐保留其语义。在多个基准和真实机械臂上提升成功率与泛化。
@misc{kim2026spark,
title = {{SPARK: Simple Post-training for Adapting pRetrained Knowledge to Robot Control}},
author = {Taekyung Kim and Jeongeun Park and Byeongho Heo and Sungjoon Choi and Sangdoo Yun and Dongyoon Han},
year = {2026},
eprint = {2611.12102},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.12102},
}
Reuses demos across arms by aligning a source encoder to a frozen target encoder on sim-rendered paired views, no pixel translation. Up to 93% zero-shot in sim; real xArm→Franka 27/30 with 10 demos per task.用仿真渲染的成对视角，把源机器人的视觉编码器对齐到冻结的目标编码器，无需像素级图像翻译即可跨机械臂复用示范。仿真零样本成功率最高 93%；真实 xArm→Franka 每任务仅用 10 条目标示范微调即 27/30 成功。
@misc{ding2026unilatent,
title = {{UniLatent: Cross-Embodiment Transfer via Latent Observation Alignment}},
author = {Yufei Ding and Yi Du and Weiduo Yuan and Hao Jiang and Dylan Goetting and Lawrence Yunliang Chen and Daniel Seita and Yue Wang},
year = {2026},
eprint = {2611.10061},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.10061},
}
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},
}
CUE (CoRL 2026 Late Breaking Results): an agent identifies friction and restitution by choosing interactions where its simulation is least accurate; on billiards it reports 72% lower held-out position error.CUE（CoRL 2026 Late Breaking Results）：智能体主动选择仿真最不准的交互来辨识摩擦和恢复系数；在台球任务上报告留出位置误差比最强基线低 72%，并部署到真实球桌。
@misc{zhang2026closing,
title = {{Closing the Dynamics Gap through Self-Directed Robotic Interaction}},
author = {Xilun Zhang and Liam Achenbach and Luis Pabon and Matthew Foutter and Marco Pavone},
year = {2026},
eprint = {2611.02691},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.02691},
}
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},
}
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},
}
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},
}
A desktop testbed under $2,500 for adversarial multi-agent RL, built around the tabletop game KLASK: open CAD and BOM, an Isaac Lab digital twin, and a sim-to-real baseline with a learned actuator model.基于桌游 KLASK 的桌面级对抗多智能体强化学习平台，成本低于 2500 美元：开放 CAD 与物料清单，提供 Isaac Lab 数字孪生，以及采用学习型执行器模型的 sim-to-real 基线。
@misc{ramachandran2026klasktron,
title = {{KlaskTron: An Open-Source Platform for Physical Adversarial Multi-Agent RL}},
author = {Aswin Ramachandran and Jona Schulz and Maurus Derungs and Carlo Angelini and Tobias Meier and Raffaello D'Andrea},
year = {2026},
eprint = {2611.03138},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.03138},
}
Reuses a frozen multi-robot value function as a differentiable surrogate for robot design, avoiding a new co-design RL loop per robot; handles over 1100 design parameters.把冻结的多机器人价值函数作为可微代理模型来优化机器人设计，无需为每台机器人重跑协同设计 RL；可处理 1100 多个设计参数。
@misc{bohlinger2026shape,
title = {{Shape Your Body: Value Gradients for Multi-Embodiment Robot Design}},
author = {Nico Bohlinger and Jan Peters},
year = {2026},
eprint = {2606.00702},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.00702},
}
Pretrains a task-agnostic latent action space in a low-fidelity simulator so real-world RL on high-DoF mobile manipulators becomes safe and efficient.在低保真仿真器中预训练与任务无关的潜在动作空间，让高自由度移动操作机器人的真实世界强化学习安全高效。
@misc{hu2025slac,
title = {{SLAC: Safe and Efficient Real-Robot Reinforcement Learning via Unsupervised Simulation Pre-Training}},
author = {Jiaheng Hu and Peter Stone and Roberto Martín-Martín},
year = {2025},
eprint = {2506.04147},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2506.04147},
}
A slow latent diffusion policy plans action chunks while a fast policy reacts to tactile/force feedback; reports >35% higher scores than visual imitation baselines on three contact-rich tasks.慢速扩散策略在隐空间规划动作块，快速策略根据触觉/力反馈高频调整；论文报告在三个接触丰富任务上比纯视觉模仿学习基线总分高 35% 以上。
A promptable humanoid foundation model: a shared latent space embeds motions, goals and rewards, so one policy can be prompted for different tasks on a real robot.可提示的人形基础模型：用共享隐空间嵌入动作、目标和奖励，同一策略可在真机上按提示执行不同任务。
@misc{li2025bfm,
title = {{BFM-Zero: A Promptable Behavioral Foundation Model for Humanoid Control Using Unsupervised Reinforcement Learning}},
author = {Yitang Li and Zhengyi Luo and Tonghe Zhang and Cunxi Dai and Anssi Kanervisto and Andrea Tirinzoni and Haoyang Weng and Kris Kitani and Mateusz Guzek and Ahmed Touati and Alessandro Lazaric and Matteo Pirotta and Guanya Shi},
year = {2025},
eprint = {2511.04131},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2511.04131},
}
Fine-tunes video world models on a robot to dream new demos, labels them with latent actions or an inverse dynamics model, and trains policies on them; a humanoid learns 20+ new behaviors from one-scene teleop data.把视频世界模型微调到目标机器人上“做梦”生成新示范，再用隐动作或逆动力学模型补上动作标签用于训练策略；人形机器人仅凭单一场景的遥操作数据学会 20 多种新行为。
@misc{jang2025dreamgen,
title = {{DreamGen: Unlocking Generalization in Robot Learning through Video World Models}},
author = {Joel Jang and Seonghyeon Ye and Zongyu Lin and Jiannan Xiang and Johan Bjorck and Yu Fang and Fengyuan Hu and Spencer Huang and Kaushil Kundalia and Yen-Chen Lin and Loic Magne and Ajay Mandlekar and Avnish Narayan and You Liang Tan and Guanzhi Wang and Jing Wang and Qi Wang and Yinzhen Xu and Xiaohui Zeng and Kaiyuan Zheng and Ruijie Zheng and Ming-Yu Liu and Luke Zettlemoyer and Dieter Fox and Jan Kautz and Scott Reed and Yuke Zhu and Linxi Fan},
year = {2025},
eprint = {2505.12705},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2505.12705},
}
DSRL — adapts a frozen diffusion policy by running RL over its input noise, not its weights; needs only forward passes, works with π0, and reports real-robot success rising from 20% to 90% in under 50 online episodes.DSRL：不改扩散策略的权重，而是对其输入的初始噪声做强化学习来"引导"输出动作；只需前向推理、可用于 π0，论文报告真机任务在不到 50 次在线交互内成功率从 20% 升至 90%。
@misc{wagenmaker2025steering,
title = {{Steering Your Diffusion Policy with Latent Space Reinforcement Learning}},
author = {Andrew Wagenmaker and Mitsuhiko Nakamoto and Yunchu Zhang and Seohong Park and Waleed Yagoub and Anusha Nagabandi and Abhishek Gupta and Sergey Levine},
year = {2025},
eprint = {2506.15799},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2506.15799},
}
FQL — a flow-matching BC policy plus a one-step policy that maximizes Q while distilling from it; no backprop through flow steps, strong on 73 OGBench / D4RL offline and offline-to-online tasks.FQL：先用行为克隆训练流匹配策略，再训练一个最大化 Q 值、同时向流策略蒸馏的单步策略；避免对多步流采样反向传播，在 73 个 OGBench / D4RL 离线及离线转在线任务上表现突出。
Tests 'embodiment scaling laws' on about 1,000 generated robots: does training on more bodies improve generalization to unseen ones?在约 1000 个生成的机器人形态上检验“形态缩放定律”：用更多形态训练能否提升对未见形态的泛化？
@misc{ai2025towards,
title = {{Towards Embodiment Scaling Laws in Robot Locomotion}},
author = {Bo Ai and Liu Dai and Nico Bohlinger and Dichen Li and Tongzhou Mu and Zhanxin Wu and K. Fay and Henrik I. Christensen and Jan Peters and Hao Su},
year = {2025},
eprint = {2505.05753},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2505.05753},
}
A simple recipe that trains on both real and simulated data to supplement expensive real-world collection for vision-based manipulation.简单的联合训练做法：同时使用真实和仿真数据，弥补视觉操作中真实数据采集昂贵的问题。
@misc{maddukuri2025sim,
title = {{Sim-and-Real Co-Training: A Simple Recipe for Vision-Based Robotic Manipulation}},
author = {Abhiram Maddukuri and Zhenyu Jiang and Lawrence Yunliang Chen and Soroush Nasiriany and Yuqi Xie and Yu Fang and Wenqi Huang and Zu Wang and Zhenjia Xu and Nikita Chernyadev and Scott Reed and Ken Goldberg and Ajay Mandlekar and Linxi Fan and Yuke Zhu},
year = {2025},
eprint = {2503.24361},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2503.24361},
}
A flow-matching policy and an action-conditioned dynamics model co-sample an action and its predicted future until they agree. Reports 56.5% (sim) and 33.3% (real) relative gains over the strongest baselines.流匹配策略与动作条件动力学模型联合采样动作及其未来观测，相互修正直到二者自洽。报告相对最强基线在仿真提升 56.5%、真机提升 33.3%。
@misc{lee2025learning,
title = {{Learning Robust Visuomotor Policies via Dynamics Alignment}},
author = {Dohyeok Lee and Jung Min Lee and Munkyung Kim and Seokhun Ju and Jin Woo Koo and Kyungjae Lee and Dohyeong Kim and Taehyun Cho and Tapomayukh Bhattacharjee and Jungwoo Lee},
year = {2025},
eprint = {2510.27114},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2510.27114},
}
Finds that shift equivariance, local receptive fields and inference choices let diffusion planners stitch sub-trajectories without TD learning; locality matters most. Proposes the Eq-Net architecture.发现平移等变性、局部感受野和推理设置使扩散规划器无需 TD 学习就能拼接子轨迹，其中局部性最关键；据此提出 Eq-Net 架构。
@misc{clark2025what,
title = {{What Do You Need for Compositional Generalization in Diffusion Planning?}},
author = {Quentin Clark and Florian Shkurti},
year = {2025},
eprint = {2505.18083},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2505.18083},
}
Exoskeleton demos are converted into pseudo-robot demonstrations; the paper reports its RISE-2 policy trained on them alone, with no robot data, performs comparably to one trained on teleoperated robot data.将外骨骼示教转换为“伪机器人”示教；论文报告仅用这些数据（无任何机器人数据）训练的 RISE-2 策略，效果与用遥操作机器人数据训练的策略相当。
@misc{fang2025airexo,
title = {{AirExo-2: Scaling up Generalizable Robotic Imitation Learning with Low-Cost Exoskeletons}},
author = {Hongjie Fang and Chenxi Wang and Yiming Wang and Jingjing Chen and Shangning Xia and Jun Lv and Zihao He and Xiyan Yi and Yunhan Guo and Xinyu Zhan and Lixin Yang and Weiming Wang and Cewu Lu and Hao-Shu Fang},
year = {2025},
eprint = {2503.03081},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2503.03081},
}
HPT: per-robot stems turn proprioception and vision into a few tokens for one shared transformer trunk, pretrained on 52 datasets (robot, sim, human video); reports 20%+ gains on unseen tasks.HPT：为每种机器人设计“输入头”，把本体感知与视觉映射成短 token 序列，送入共享的 Transformer 主干；在 52 个数据集（真机、仿真、人类视频）上预训练，报告在未见任务上微调提升 20% 以上。
LAPA pretrains a VLA without action labels: it learns discrete latent actions between video frames, predicts them from image + text, then maps them to robot actions; reports beating OpenVLA on real tasks.LAPA 无需动作标注即可预训练 VLA：先从视频相邻帧间学习离散隐动作，再训练模型由图像和指令预测隐动作，最后用少量机器人数据映射到真实动作；报告在真机任务上优于 OpenVLA。
@misc{ye2024latent,
title = {{Latent Action Pretraining from Videos}},
author = {Seonghyeon Ye and Joel Jang and Byeongguk Jeon and Sejune Joo and Jianwei Yang and Baolin Peng and Ajay Mandlekar and Reuben Tan and Yu-Wei Chao and Bill Yuchen Lin and Lars Liden and Kimin Lee and Jianfeng Gao and Luke Zettlemoyer and Dieter Fox and Minjoon Seo},
year = {2024},
eprint = {2410.11758},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2410.11758},
}
Treats egocentric human video with 3D hand tracking (Project Aria glasses) as robot data and co-trains one policy on both; reports an extra hour of human hand data helps more than an extra hour of robot data.把佩戴 Project Aria 眼镜采集、带三维手部跟踪的第一视角人类视频当作机器人数据，与机器人数据共同训练同一个策略；论文报告多 1 小时人手数据比多 1 小时机器人数据更有用。
@misc{kareer2024egomimic,
title = {{EgoMimic: Scaling Imitation Learning via Egocentric Video}},
author = {Simar Kareer and Dhruv Patel and Ryan Punamiya and Pranay Mathur and Shuo Cheng and Chen Wang and Judy Hoffman and Danfei Xu},
year = {2024},
eprint = {2410.24221},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2410.24221},
}
@misc{bohlinger2024one,
title = {{One Policy to Run Them All: an End-to-end Learning Approach to Multi-Embodiment Locomotion}},
author = {Nico Bohlinger and Grzegorz Czechmanowski and Maciej Krupka and Piotr Kicki and Krzysztof Walas and Jan Peters and Davide Tateo},
year = {2024},
eprint = {2409.06366},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2409.06366},
}
An open-source stack for training manipulation policies with RL on a real arm — learned reward classifiers, reset-free training and an impedance controller.在真实机械臂上用强化学习训练操作策略的开源软件栈：学习式奖励分类器、免重置训练与阻抗控制器。
@misc{luo2024serl,
title = {{SERL: A Software Suite for Sample-Efficient Robotic Reinforcement Learning}},
author = {Jianlan Luo and Zheyuan Hu and Charles Xu and You Liang Tan and Jacob Berg and Archit Sharma and Stefan Schaal and Chelsea Finn and Abhishek Gupta and Sergey Levine},
year = {2024},
eprint = {2401.16013},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2401.16013},
}
DPPO — fine-tunes a pre-trained diffusion policy with PPO by treating the denoising steps as part of the MDP; reports stable gains over other RL fine-tuning methods and sim-to-real transfer.DPPO：把去噪步骤视为 MDP 的一部分，用 PPO 微调预训练扩散策略；报告称比其他强化学习微调方法更稳定，并实现了仿真到真机迁移。
@misc{ren2024diffusion,
title = {{Diffusion Policy Policy Optimization}},
author = {Allen Z. Ren and Justin Lidard and Lars L. Ankile and Anthony Simeonov and Pulkit Agrawal and Anirudha Majumdar and Benjamin Burchfiel and Hongkai Dai and Max Simchowitz},
year = {2024},
eprint = {2409.00588},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2409.00588},
}
@misc{mccarthy2024towards,
title = {{Towards Generalist Robot Learning from Internet Video: A Survey}},
author = {Robert McCarthy and Daniel C. H. Tan and Dominik Schmidt and Fernando Acero and Nathan Herr and Yilun Du and Thomas G. Thuruthel and Zhibin Li},
year = {2024},
eprint = {2404.19664},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2404.19664},
}
PIRLNav — behavior cloning on human ObjectNav demonstrations, then RL fine-tuning of the same policy; the paper reports 65.0% success on Habitat ObjectNav, 5 points above the previous best.PIRLNav：先在人类 ObjectNav 演示上做行为克隆，再用强化学习微调同一策略；论文报告在 Habitat ObjectNav 上成功率 65.0%，比此前最佳高 5 个百分点。
@misc{ramrakhya2023pirlnav,
title = {{PIRLNav: Pretraining with Imitation and RL Finetuning for ObjectNav}},
author = {Ram Ramrakhya and Dhruv Batra and Erik Wijmans and Abhishek Das},
year = {2023},
eprint = {2301.07302},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2301.07302},
}
RLPD — a few simple changes to off-policy RL (symmetric sampling of offline and online data, layer norm, ensembles) make prior data speed up online learning.RLPD：对离策略强化学习做几处简单改动（离线/在线数据对称采样、层归一化、集成），让已有数据加速在线学习。
@misc{ball2023efficient,
title = {{Efficient Online Reinforcement Learning with Offline Data}},
author = {Philip J. Ball and Laura Smith and Ilya Kostrikov and Sergey Levine},
year = {2023},
eprint = {2302.02948},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2302.02948},
}
Keeps conservative offline Q-values from dropping below a reference policy's value, which avoids the early performance dip when an offline-pretrained policy starts online fine-tuning.约束保守的离线 Q 值不低于参考策略的价值，从而避免离线预训练策略开始在线微调时常见的性能先降后升。
@misc{nakamoto2023cal,
title = {{Cal-QL: Calibrated Offline RL Pre-Training for Efficient Online Fine-Tuning}},
author = {Mitsuhiko Nakamoto and Yuexiang Zhai and Anikait Singh and Max Sobol Mark and Yi Ma and Chelsea Finn and Aviral Kumar and Sergey Levine},
year = {2023},
eprint = {2303.05479},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2303.05479},
}
Discretizes each action dimension and predicts Q-values per dimension autoregressively with a Transformer, plus a conservative regularizer; trained offline on large real robot datasets mixing demos and autonomous data.将每个动作维度离散化，用 Transformer 自回归地逐维预测 Q 值，并加入保守正则；在混合演示与自主采集数据的大规模真实机器人数据集上离线训练。
@misc{chebotar2023q,
title = {{Q-Transformer: Scalable Offline Reinforcement Learning via Autoregressive Q-Functions}},
author = {Yevgen Chebotar and Quan Vuong and Alex Irpan and Karol Hausman and Fei Xia and Yao Lu and Aviral Kumar and Tianhe Yu and Alexander Herzog and Karl Pertsch and Keerthana Gopalakrishnan and Julian Ibarz and Ofir Nachum and Sumedh Sontakke and Grecia Salazar and Huong T Tran and Jodilyn Peralta and Clayton Tan and Deeksha Manjunath and Jaspiar Singht and Brianna Zitkovich and Tomas Jackson and Kanishka Rao and Chelsea Finn and Sergey Levine},
year = {2023},
eprint = {2309.10150},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2309.10150},
}
LLM+P has an LLM translate problems into a formal form for classical planners, which handle the long-horizon planning LLMs struggle with.LLM+P 让大模型把问题转成规范形式交给经典规划器，解决大模型不擅长的长程规划。
@misc{liu2023llm,
title = {{LLM+P: Empowering Large Language Models with Optimal Planning Proficiency}},
author = {Bo Liu and Yuqian Jiang and Xiaohan Zhang and Qiang Liu and Shiqi Zhang and Joydeep Biswas and Peter Stone},
year = {2023},
eprint = {2304.11477},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2304.11477},
}
Open-source tactile RL environments with sim-to-real transfer across three low-cost optical sensors (TacTip, DIGIT, DigiTac), for comparing them on the same tasks.开源触觉强化学习环境，支持三种低成本光学触觉传感器（TacTip、DIGIT、DigiTac）的仿真到真机迁移，便于在相同任务上比较。
@misc{lin2022tactile,
title = {{Tactile Gym 2.0: Sim-to-real Deep Reinforcement Learning for Comparing Low-cost High-Resolution Robot Touch}},
author = {Yijiong Lin and John Lloyd and Alex Church and Nathan F. Lepora},
year = {2022},
eprint = {2207.10763},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2207.10763},
}
BeT — clusters actions into k bins plus a continuous offset so a transformer can imitate multimodal demonstrations instead of averaging them.BeT：把动作聚成 k 个类别再加连续偏移，让 Transformer 能模仿多模态示教，而不是把不同模式平均掉。
@misc{shafiullah2022behavior,
title = {{Behavior Transformers: Cloning $k$ modes with one stone}},
author = {Nur Muhammad Mahi Shafiullah and Zichen Jeff Cui and Ariuntuya Altanzaya and Lerrel Pinto},
year = {2022},
eprint = {2206.11251},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2206.11251},
}
Early use of LLMs as few-shot planners for embodied agents following natural-language instructions in visual environments.较早把大模型用作具身智能体的少样本规划器，在视觉环境中执行自然语言指令。
@misc{song2022llm,
title = {{LLM-Planner: Few-Shot Grounded Planning for Embodied Agents with Large Language Models}},
author = {Chan Hee Song and Jiaman Wu and Clayton Washington and Brian M. Sadler and Wei-Lun Chao and Yu Su},
year = {2022},
eprint = {2212.04088},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2212.04088},
}
A discriminator trained on unstructured mocap clips acts as a style reward, so RL learns tasks with natural motion without hand-picked reference clips; widely reused as a gait prior for legged robots.用在无结构动捕片段上训练的判别器作为"风格奖励"，强化学习无需逐段指定参考动作即可学到自然的动作；后来被广泛用作足式机器人的步态先验。
@misc{peng2021amp,
title = {{AMP: Adversarial Motion Priors for Stylized Physics-Based Character Control}},
author = {Xue Bin Peng and Ze Ma and Pieter Abbeel and Sergey Levine and Angjoo Kanazawa},
year = {2021},
eprint = {2104.02180},
archivePrefix = {arXiv},
primaryClass = {cs.GR},
url = {https://arxiv.org/abs/2104.02180},
}
IQL — learns values with expectile regression without ever querying actions outside the dataset, then extracts a policy by advantage-weighted regression; a common base for offline-to-online robot RL.IQL：用期望分位数回归学习价值函数、从不查询数据集外的动作，再用优势加权回归提取策略；是机器人离线到在线强化学习的常用基础。
Represents the policy as an energy-based model over actions instead of a regressor, which handles multimodal and discontinuous demonstrations much better.用基于能量的模型（而非直接回归）表示策略，更好地处理多模态、不连续的示范数据。
@misc{florence2021implicit,
title = {{Implicit Behavioral Cloning}},
author = {Pete Florence and Corey Lynch and Andy Zeng and Oscar Ramirez and Ayzaan Wahid and Laura Downs and Adrian Wong and Johnny Lee and Igor Mordatch and Jonathan Tompson},
year = {2021},
eprint = {2109.00137},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2109.00137},
}
DrQ-v2 - a simple model-free actor-critic with image augmentation that learns continuous control from pixels (incl. humanoid tasks in the DeepMind Control Suite) on a single GPU; a common visual-RL baseline.DrQ-v2：带图像增强的简单无模型 actor-critic，单 GPU 即可从像素学习连续控制（包括 DeepMind Control Suite 的人形任务），是常用的视觉强化学习基线。
CQL — penalizes Q-values on actions outside the dataset so the learned value is a lower bound; a standard offline-RL baseline for learning from fixed robot logs.CQL：压低数据集外动作的 Q 值，使学到的价值成为下界；是从固定机器人日志中学习的标准离线强化学习基线。
Advantage-weighted actor-critic that pre-trains on prior demonstrations and keeps improving online, shown on real dexterous-hand tasks.优势加权的 Actor-Critic：先用已有示范数据预训练，再在线持续提升，并在真实灵巧手任务上验证。
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},
}
Learns a compact vision + force/torque representation with self-supervised objectives, then trains a peg-insertion policy with RL on top of it on a real robot.用自监督目标学习视觉 + 力/力矩的紧凑联合表征，再在其上用强化学习在真实机器人上训练插孔策略。
@misc{lee2018making,
title = {{Making Sense of Vision and Touch: Self-Supervised Learning of Multimodal Representations for Contact-Rich Tasks}},
author = {Michelle A. Lee and Yuke Zhu and Krishnan Srinivasan and Parth Shah and Silvio Savarese and Li Fei-Fei and Animesh Garg and Jeannette Bohg},
year = {2018},
eprint = {1810.10191},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/1810.10191},
}
Off-policy RL with an entropy bonus; sample-efficient enough to become the default for learning directly on real robots.带熵奖励的离策略强化学习；样本效率较高，成为真机上直接学习的常用默认算法。
@misc{haarnoja2018soft,
title = {{Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor}},
author = {Tuomas Haarnoja and Aurick Zhou and Pieter Abbeel and Sergey Levine},
year = {2018},
eprint = {1801.01290},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/1801.01290},
}
Off-policy Q-learning with a cross-entropy-method action optimizer, trained on several hundred thousand real grasp attempts collected by a fleet of robots; reports 96% grasp success on unseen objects.使用交叉熵方法优化动作的离策略 Q 学习，在机器人集群采集的数十万次真实抓取上训练；报告对未见物体的抓取成功率为 96%。
@misc{kalashnikov2018qt,
title = {{QT-Opt: Scalable Deep Reinforcement Learning for Vision-Based Robotic Manipulation}},
author = {Dmitry Kalashnikov and Alex Irpan and Peter Pastor and Julian Ibarz and Alexander Herzog and Eric Jang and Deirdre Quillen and Ethan Holly and Mrinal Kalakrishnan and Vincent Vanhoucke and Sergey Levine},
year = {2018},
eprint = {1806.10293},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/1806.10293},
}
GAIL — learns a policy whose state-action distribution a discriminator cannot tell apart from the expert's, skipping explicit reward recovery.GAIL：训练策略使判别器无法区分其状态-动作分布与专家数据，无需显式恢复奖励函数。
DAgger — query the expert on the states the learner itself visits, fixing the compounding errors of plain behavior cloning.DAgger：在学习者自己访问到的状态上向专家查询标注，解决单纯行为克隆的误差累积问题。
@misc{ross2010reduction,
title = {{A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning}},
author = {Stephane Ross and Geoffrey J. Gordon and J. Andrew Bagnell},
year = {2010},
eprint = {1011.0686},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/1011.0686},
}
George Nehma, Patrick Quinn, Madhur Tiwari, Camilo Riano-Rios
Abstract摘要
Soft powered descent landings are becoming evermore important in the field of space exploration and scientific advancement, because of their ability to deliver mission critical payloads, both highly accurately and safely. The high efficiency restriction on planetary landings due to the severely limited fuel supply results in the need for optimal guidance trajectories and control. Model based reinforcement learning is proving to be an important tool in simplifying the highly complex process of generating optimal guidance trajectories and control. In this work, we present DreamLander, a state …
@misc{nehma2026dreamlander,
title = {{DreamLander: A 6-DOF World Model based Reinforcement Learning Planetary Lander}},
author = {George Nehma and Patrick Quinn and Madhur Tiwari and Camilo Riano-Rios},
year = {2026},
eprint = {2611.12824},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.12824},
}
Omer Hemo, Ido Sefi, Mirco Mutti, Aviv Tamar et al.
Abstract摘要
Reinforcement Learning (RL) algorithms often rely on data collected from a target environment, which can be expensive and noisy. Although simulators can generate large amounts of additional data, standard simulation mixing approaches, which directly combine simulated and real experience, can degrade performance due to simulator inaccuracies, especially when they bias the solution away from the optimal policy. We propose a conservative simulation-aided learning method based on control variates, where simulated data is used solely to reduce variance in the learning process rather than directly …
@misc{hemo2026simulation,
title = {{Simulation-aided Reinforcement Learning with Control Variates}},
author = {Omer Hemo and Ido Sefi and Mirco Mutti and Aviv Tamar and Kfir Y. Levy},
year = {2026},
eprint = {2611.12643},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2611.12643},
}
Meraj Mammadov, Pedro Zuidberg Dos Martires, Johannes Andreas Stork
Abstract摘要
Reinforcement learning (RL) from raw sensory inputs such as images or onboard sensors can be challenging due to high-dimensional observations and sparse rewards. A common strategy is to train a teacher policy with access to privileged state information and then distill it into a student policy that acts from raw inputs alone. However, when the teacher relies on information unavailable to the student, exact imitation may be impossible, creating an irreducible imitation gap. Existing approaches typically mitigate this mismatch through careful reward shaping or additional reinforcement learning …
@misc{mammadov2026rail,
title = {{RAIL: Representation-Aligned Imitation Learning for Student-Compatible Teacher Policies}},
author = {Meraj Mammadov and Pedro Zuidberg Dos Martires and Johannes Andreas Stork},
year = {2026},
eprint = {2611.12251},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2611.12251},
}
In patrol missions involving multiple unmanned surface vehicles (USVs), a designated command USV can support fleet management by sequentially rendezvousing with each patrolling USV. These operations require motion-feasible route planning and priority-aware rendezvous sequencing, with the objective of minimizing mission completion time and total tardiness relative to priority-dependent due dates. We propose a framework combining grid-based path planning, surrogate travel-time prediction, and graph-based reinforcement learning. The path planner accounts for ocean currents and approximates …
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},
}
Generalist robot policies often fail at one step of a task they can otherwise perform. We ask whether such a failure can be repaired by editing the policy's weights directly, without gradients, training data, or a change to the instruction. Told to put the cube then the banana in the bowl, a zero-shot $π_{0.5}$ vision-language-action policy in RoboLab releases the cube before it reaches the bowl. Replaying 21 of these decisions under rewordings shows that it binds "in the bowl" to the wrong object. Closed-form instruction distillation copies the per-layer effect of an explicit rewording into …
Exploration in deep reinforcement learning (RL) for continuous control is commonly driven by temporally uncorrelated Gaussian (white) action noise, which leads to jerky motions with locally constrained exploration. In contrast, temporally correlated coloured noise, particularly pink noise, has been shown to support more efficient exploration in RL. Motivated by these findings, we inquire whether scheduling changes to the noise colour over the course of training can outperform white noise for continuous control robotics. We perform a systematic study of 49 linear schedules between white and …
Roshan S. Kaundinya, Hehui Zheng, Robert K. Katzschmann, George Haller
Abstract摘要
Reinforcement learning (RL) requires large amounts of interaction data or a reliable simulator, both hard to obtain for nonlinear soft robotic hardware. Recent work shows that a time-scale separation inherent to soft robot dynamics justifies their reduction to low-dimensional invariant manifolds, known as adiabatic spectral submanifolds (aSSMs). We demonstrate that an aSSM-reduced model, a simple polynomial ordinary differential equation learned purely from pre-collected data, can serve as an RL training environment. On a pneumatically actuated soft arm, we train a trajectory-tracking policy …
@misc{kaundinya2026accelerated,
title = {{Accelerated Reinforcement Learning for Soft Robotics: Hours Instead of Years}},
author = {Roshan S. Kaundinya and Hehui Zheng and Robert K. Katzschmann and George Haller},
year = {2026},
eprint = {2611.11718},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.11718},
}
The CoRL 2026 RoboEval challenge has eight bimanual tasks, and it ranks a policy not only on whether it succeeds but on how: every stage of the task, done safely, efficiently and with both arms in coordination, scored against human demonstrations. My solution has two parts. The generic part, carried over from my LeHome challenge solution, is off-policy RL on a $π_{0.5}$-based policy that is also its own value function. It is trained with AWR and RECAP on advantages of the exact official score, in a loop where every rollout is at the same time a Thompson-sampling trial of about 20 inference …
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},
}
Jaeyoun Choi, Niklas Schmid, Oswin So, Augustinos Saravanos et al.
Abstract摘要
Disturbances encountered during autonomous navigation generally vary across space, often discontinuously, as obstacles redirect wind and changing terrain alters traction. Tracking controllers typically assume sufficient control authority to mitigate disturbances and maintain safety; the regime where this assumption fails remains underexplored. We present Belief-guided Residual Adaptive Control (BRACE), a hierarchical reinforcement learning (RL) framework for reach-avoid navigation under unknown, bounded, spatially varying disturbances. BRACE infers a disturbance belief from model prediction …
@misc{choi2026brace,
title = {{BRACE: Belief-guided Residual Adaptive Control for Safe Navigation under Unknown Disturbances}},
author = {Jaeyoun Choi and Niklas Schmid and Oswin So and Augustinos Saravanos and Chuchu Fan},
year = {2026},
eprint = {2611.10938},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.10938},
}
Kevin Park, Izat Rasyad, Dvij Kalaria, Shankar Sastry
Abstract摘要
Training vision-language-action (VLA) policies typically depends on substantial robot demonstration data, making human teleoperation a costly and cumbersome bottleneck, particularly for whole-body humanoid interaction. Reinforcement learning in simulation offers an alternative for scaling this data, but designing task-specific rewards remains challenging. To address this, we propose DreamVLA, which uses generated motion trajectories to guide reinforcement learning (RL) in simulation, converting them into reliable task execution trajectories. These trajectories train VLA models in simulation …
@misc{park2026dreamvla,
title = {{DreamVLA: From Generated Human Motions to Humanoid Vision-Language-Action Policies}},
author = {Kevin Park and Izat Rasyad and Dvij Kalaria and Shankar Sastry},
year = {2026},
eprint = {2611.10872},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.10872},
}
Iterative neural policies usually spend the same computation on every decision, although the benefit of additional processing varies across states. We introduce Adaptive Internal Computation (AIC), an iterative policy that learns when to stop refining its latent representation and act. In AIC, we train action selection and stopping jointly, an expected-depth penalty values computation and intermediate value predictions provide auxiliary supervision. We show that when only some states benefit from further refinement, an ideal state-dependent allocator outperforms every state-independent …
Ahmed Zeer, Osama Shbib, Muzaffer Kaan Yuce, Himmet Toprak Kesgin et al.
Abstract摘要
Search allows a robotic policy to evaluate candidate futures before committing to an action. Motivated by test-time planning for vision-language action (VLA) models, we study selective search over action chunks in PushT using a frozen Action Chunking with Transformers (ACT) policy. Beam search generates perturbed alternatives around the policy prediction and scores them through rollouts from cloned simulator states. Across 50 evaluation episodes, beam search raises task success from 0% to 18% and increases average cumulative reward from 39.44 to 103.2. We introduce One Step Ahead (OSA), a …
@misc{zeer2026look,
title = {{Look Ahead: Search-Driven Reasoning in Embodied AI}},
author = {Ahmed Zeer and Osama Shbib and Muzaffer Kaan Yuce and Himmet Toprak Kesgin and Mehmet Fatih Amasyali},
year = {2026},
eprint = {2611.10769},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2611.10769},
}
Reliable navigation with a single monocular limited field-of-view camera is challenging because the lack of depth measurements and limited spatial awareness complicate collision checking. Existing approaches to tackle these challenges fall into two categories. Model-based planners can incorporate explicit collision-avoidance guidance which leads to better generalization but are often computationally expensive and require manual tuning of the algorithmic parameters. In contrast, end-to-end approaches trained by imitation learning can be flexible but typically encode safety only implicitly and …
Rongqian Chen, Yiding Chen, Jijie Zhou, Tom Z. Jiahao
Abstract摘要
Most robotic handwriting rigidly mounts the pen and reduces writing to trajectory tracking. A freely held pen instead couples grasp stability, tip motion, and contact force. We present Grip, Trace, Write, a force-aware simulation environment and learning framework for high-DoF handwriting. A two-branch actor separates pen retention from arm-wrist writing while preserving their closed-loop coupling. Training learns a grasp controller, initializes writing from inverse-kinematics demonstrations by behavior cloning, and improves writing through reinforcement learning with rewards on ink, force, …
@misc{chen2026grip,
title = {{Grip, Trace, Write: Force-Aware Imitation and Reinforcement Learning for High-DoF Robotic Handwriting}},
author = {Rongqian Chen and Yiding Chen and Jijie Zhou and Tom Z. Jiahao},
year = {2026},
eprint = {2611.10451},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.10451},
}
Phillip J. K. Christoffersen, Rachel Ma, Patrick Barragán, Dylan Hadfield-Menell
Abstract摘要
In this work, we train agents to act under uncertainty according to specified, belief-dependent rules. Deep reinforcement learning (RL) can in theory train agents to follow rules if encoded into well-designed rewards, but in practice, RL fails to perfectly optimize these rewards, leading to ineffective or noncompliant agent behavior. Belief-space planners by construction produce behavior in line with developer specification, but require hand-programming online planners and belief-state estimators, adding significant engineering complexity at scale. We combine the two: we plan offline to …
@misc{christoffersen2026deploying,
title = {{Deploying Physically and Epistemically Compliant Agents by Distilling Belief-Space Task and Motion Planners}},
author = {Phillip J. K. Christoffersen and Rachel Ma and Patrick Barragán and Dylan Hadfield-Menell},
year = {2026},
eprint = {2611.10430},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.10430},
}
Visuomotor policies can handle many robot manipulation tasks, but understanding their failures remains difficult. Unfamiliar conditions and small action errors can lead to failure, yet watching the robot alone tells us little about what went wrong inside the policy. We introduce MIND (Mechanistic Interpretability for Neural Diagnosis), a method that looks inside a trained policy to detect failures during execution and identify the internal concepts involved. MIND uses a sparse autoencoder to break down the policy's internal activations into features that can be studied individually. We give …
@misc{ahmed2026detecting,
title = {{Detecting and Explaining Robot Policy Failures from Sparse Internal Features}},
author = {Riad Ahmed and Sujosh Nag and Moniruzzaman Akash and Momotaz Begum},
year = {2026},
eprint = {2611.10400},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.10400},
}
Aryaman Gupta, Xiaoyue Wu, Rana Soltani Zarrin, Mark Van der Merwe
Abstract摘要
Generative robot policies based on diffusion and flow matching can model complex, multimodal behaviors, but remain susceptible to covariate shift during closed-loop execution. Existing approaches mitigate such failures through additional demonstrations, policy fine-tuning, or test-time guidance, but often require human intervention, engineered rewards, or other task-specific signals. We introduce GUARD, an anomaly-informed Q-steering framework that leverages the policy's training distribution to construct a dense, failure-anticipatory steering signal. GUARD pairs sparse task rewards with …
@misc{gupta2026guard,
title = {{GUARD: Anomaly-Informed Q-Steering for Generative Robot Policies in Dexterous Manipulation}},
author = {Aryaman Gupta and Xiaoyue Wu and Rana Soltani Zarrin and Mark Van der Merwe},
year = {2026},
eprint = {2611.10343},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.10343},
}
A robot that keeps learning after deployment is corrected by several people in turn, and one of them may later ask to be removed. We show that whether the robot can forget that person depends on how people correct it. The goal of removal is the robot that would exist if the person had never been there, which we compute exactly in two simulated tasks with scripted users. When users correct the robot by taking over control, an ordered log of training updates is enough: retraining on the remaining sessions in their original order gives a robot as close to the goal as training noise allows. …
Jianghan Zhang, Roland Schwan, Gabriel Bravo-Palacios, John Zhang et al.
Abstract摘要
Differentiable model predictive control (MPC) enables structured, constraint-aware policies to be optimized end-to-end. However, learning MPC policies faces two bottlenecks: (1) nonsmooth sensitivities at active-set transitions and (2) expensive policy evaluation, since each requires an optimization solve. We first introduce SmoothedTurboMPC, a GPU-parallel differentiable MPC solver combining sequential quadratic programming (SQP) with a log-barrier-smoothed alternating direction method of multipliers (ADMM) inner solver. Its design enables smooth gradients, exact sensitivities at …
@misc{zhang2026efficient,
title = {{Efficient Learning of Differentiable MPC Policies on the GPU via Smoothed ADMM and First-Order Reinforcement Learning}},
author = {Jianghan Zhang and Roland Schwan and Gabriel Bravo-Palacios and John Zhang and Thomas Lew and Brian Plancher},
year = {2026},
eprint = {2611.09937},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.09937},
}
Zhaofan Zhang, Yue Chang, Rufeng Chen, Hechang Chen et al.
Abstract摘要
Ensuring safe and reliable navigation under diverse sources of uncertainty remains a central challenge for autonomous systems. Within distributional reinforcement learning (DistRL), Implicit Quantile Networks (IQN) have attracted interest for navigation under uncertainty because they enable risk adaptation through flexible quantile sampling. Existing approaches, however, typically use a fixed or randomly sampled Conditional Value-at-Risk (CVaR) level during training and adjust the risk level during deployment using hand-tuned hyperparameters or external sensor signals. This scalar-CVaR …
Chang Rajani, Luke Sebastien Marsden, Stephan Kunz, Claudio Mucignat et al.
Abstract摘要
Particle image velocimetry (PIV) reconstructs a flow field from the motion of tracer particles but the reconstruction quality is bounded by how well the flow structure of interest is resolved in the images. The acquisition position is typically fixed by the operator before the measurement. We cast camera placement as a sequential decision-making problem and train a reinforcement learning agent to control a 6-DoF manipulator equipped with a camera, enabling it to locate, frame and resolve a hydrofoil wake in a water tunnel. The agent commands the camera pose and observes the scene only through …
@misc{rajani2026autonomous,
title = {{Autonomous Particle Image Velocimetry Using Zero-Shot Sim-to-Real Reinforcement Learning}},
author = {Chang Rajani and Luke Sebastien Marsden and Stephan Kunz and Claudio Mucignat and Ivan Lunati},
year = {2026},
eprint = {2611.09679},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.09679},
}
Guangyuan Wang, Haishan Xia, Zitao Zhang, Kejia Chen et al.
Abstract摘要
Real-world robot deployment requires adapting learned policies to unseen obstacles without collecting new demonstrations or retraining. Existing policy-level guidance often applies obstacle corrections independently at individual horizon steps, which may address local obstacle violations but fail to coordinate the action sequence into a complete, task-directed detour. How obstacle responses should be propagated across the action horizon to coordinate complete, task-directed avoidance motion remains underexplored. We introduce an inferencetime obstacle-guidance framework for pretrained …
@misc{wang2026horizon,
title = {{Horizon-Coupled Obstacle Guidance for Relative-Action Diffusion Policies}},
author = {Guangyuan Wang and Haishan Xia and Zitao Zhang and Kejia Chen and Yuanchen Sun and Zhenguo Sun and Donghao Zhang and Yu Zhang and others},
year = {2026},
eprint = {2611.09340},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.09340},
}
Learning locomotion directly on a humanoid robot is a promising direction to avoid the sim-to-real gap. However, locomotion rewards in Reinforcement Learning (RL) typically track the base velocity, which cannot be measured onboard without an external motion capture system or well-calibrated state estimator. This is why we introduce \method{}, which learns humanoid locomotion from minutes of samples with a reward that only requires the onboard Inertial Measurement Unit (IMU). Instead of tracking the base velocity, STRIDE rewards the robot for following a commanded acceleration profile. To …
Paul Knaußt, Kay Pompetzki, Nico Bohlinger, Niklas Funk et al.
Abstract摘要
Geometric assembly requires fine spatiotemporal coordination to place parts without causing collisions or deadlocks under tight clearances. To address this, we present an object-centric reinforcement learning framework that treats each part as an agent and learns a shared policy for continuous assembly control. A graph neural network represents the pieces, their target poses, and their spatial relationships, and predicts translational and angular velocities for each component. Per-agent proximal policy optimization with a centralized graph critic directly learns motion and temporal …
@misc{knaut2026it,
title = {{It Takes Two to Tangram: Multi-Agent Object Centric Reinforcement Learning for Assembly}},
author = {Paul Knaußt and Kay Pompetzki and Nico Bohlinger and Niklas Funk and Cristiana de Farias and Rolf Findeisen and Jan Peters},
year = {2026},
eprint = {2611.08814},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.08814},
}
Yixiao Wang, Peiyi Chen, Mustafa Mukadam, Shreekant Gayaka et al.
Abstract摘要
Learning dexterous manipulation from limited visual demonstrations is challenging because policies must acquire both perceptual grounding and precise control, while reinforcement learning (RL) in simulation can acquire rich control behaviors from large-scale interaction with privileged information. We introduce \textbf{Latent RL Prior}, a two-stage framework that reuses this control knowledge for data-efficient visual policy learning. During privileged pretraining, we distill large-scale RL rollouts into a flow-matching action decoder conditioned on proprioception and latent task context. …
@misc{wang2026latent,
title = {{Latent RL Priors for Data-Efficient Dexterous Manipulation}},
author = {Yixiao Wang and Peiyi Chen and Mustafa Mukadam and Shreekant Gayaka and Masayoshi Tomizuka and Son Tran},
year = {2026},
eprint = {2611.08458},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.08458},
}
Online 3D bin packing with a processing buffer requires choosing which box to pack, its orientation, and its placement while keeping the stack stable under unknown future arrivals. A collapse ends the episode and forfeits the remaining volume. Existing methods use geometric rules or hard stability masks, limiting how risk can influence the joint choice of box, orientation, and placement. We propose Stability-Hybrid On-Policy Reinforcement Learning (SHORE), which combines a learned stability probability of a placement with an on-policy estimate of its return conditional on stability. A 3D …
Mateo Cl\'emente, George Z. Xue, Amin Abyaneh, Liheng Ma et al.
Abstract摘要
World-action models adapt pretrained video generators, and vision-language-action models build on pretrained vision-language backbones; both tie policy architecture and inference cost to large transformers designed for generation rather than control. It remains unclear how much these generation-oriented choices affect control. In this work, we introduce Latent Action JEPA (La-JEPA), a compact world-action model whose policy transformer is trained from scratch in the feature space of a pretrained V-JEPA 2.1 encoder. Following Fast-WAM, La-JEPA learns to predict future features jointly with …
@misc{clemente2026la,
title = {{La-JEPA: Compact Robot Policies with Latent World Modeling}},
author = {Mateo Cl\'emente and George Z. Xue and Amin Abyaneh and Liheng Ma and Rui Heng Yang and Tongtong Cao and Zhanguang Zhang and Yingxue Zhang},
year = {2026},
eprint = {2611.08156},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.08156},
}
Robot reinforcement learning increasingly runs environments outside the trainer: in another process, on another machine, in another language. Many faults at this boundary raise no error. We make the boundary a wire protocol and ask what notices a fault in it. On Pendulum, HalfCheetah, Hopper and a 7-joint arm (Pusher), we injected up to 23 faults into what crosses the wire and 2 into the trainer's log, from one value to every value. Unless every value was wrong, a learning-curve threshold caught at most one run in ten, and a comparison with ten fault-free seeds at most 21%. A clean evaluation …
Xuenmin Hu, Qingxue Qian, Wei Luo, Benyi Ma et al.
Abstract摘要
Legged robots equipped with manipulators have shown significant potential in the fields of daily life and industry in complex environments. However, existing methods mainly focus on global navigation or manipulation independently, while overlooking the last-mile navigation problem, where the robot must jointly determine the base placement, approach direction, and grasp execution to achieve reliable object grasping, and different manipulation stages require distinct control behaviors. To address these challenges, we propose Last-Mile Grasp-Navigation (LMGN), a unified policy learning model for …
@misc{hu2026lmgn,
title = {{LMGN: A Unified Policy Learning Model for Last-Mile Grasp-Navigation}},
author = {Xuenmin Hu and Qingxue Qian and Wei Luo and Benyi Ma and Bo Tang and Long Chen},
year = {2026},
eprint = {2611.07515},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.07515},
}
Robots increasingly run large vision-language-action policies on remote servers, making the camera uplink the bottleneck. Existing video codecs spend bits on content the receiver could predict: the robot's own motion. We introduce the kinematic hypothesis context, a zero-rate conditioning signal that both ends build from joint states, the robot model and the camera calibration. Ego-motion homographies, per-link robot reprojections, an end-effector plane and a background memory warp the previous reconstruction into hypotheses that condition a learned video codec. We prove that history-only and …
Yuanchu Liang, Yue Yang, An-Chi He, Hanna Kurniawati
Abstract摘要
Pre-trained diffusion and flow policies have emerged as powerful backbones for robotics, yet adapting them to novel data via reinforcement learning (RL) remains computationally expensive, sample inefficient and lacks theoretical guarantees. Prior approaches that fine-tune the initial noise bypass costly backpropagation through time but struggle with the "noise aliasing" problem: there does not exista natural link between the noise space value and the action space value. This paper resolves these challenges by introducing the Latent Noise Linear Markov Decision Process (LNL-MDP). We derive the …
@misc{liang2026efficient,
title = {{Efficient Diffusion Policy Fine Tuning with Latent Noise Representation Bridging}},
author = {Yuanchu Liang and Yue Yang and An-Chi He and Hanna Kurniawati},
year = {2026},
eprint = {2611.07275},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.07275},
}
Vinay Rodrigues, Sara Zeynalpour, Gokhan Solak, Sebastian Hjorth et al.
Abstract摘要
Deformable linear object (DLO) manipulation is challenging because high-dimensional configurations, nonlinear contact dynamics, and diverse material and geometric properties produce a broad range of object responses that are both difficult to model and data-intensive to learn. This work studies reinforcement learning for planar manipulation of a one-end-grasped DLO, focusing on training efficiency, observation design, robustness, and zero-shot sim-to-real transfer. We compare three Proximal Policy Optimization training stacks and find that the GPU-parallel MuJoCo Playground stack enables …
@misc{rodrigues2026empirical,
title = {{An Empirical Study of Reinforcement Learning for Keypoint Positioning of Deformable Linear Objects}},
author = {Vinay Rodrigues and Sara Zeynalpour and Gokhan Solak and Sebastian Hjorth and Arash Ajoudani},
year = {2026},
eprint = {2611.07261},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.07261},
}
Michal Korniak, Marco Bagatella, Ralf R\"omer, Michal Nauman et al.
Abstract摘要
The introduction of generalist robot policies has enabled a single, large model to learn to perform a variety of tasks from a vast pre-training corpus. While these models often exhibit powerful zero-shot generalization capabilities, they typically require an adaptation phase to achieve sufficient performance in deployment. Iterated self-improvement via reinforcement learning (RL) has demonstrated promise in enabling this adaptation, yet such approaches have been largely explored in single-task settings - where the goal is to simply improve on a narrow deployment task - which lies in stark …
@misc{korniak2026reinforcement,
title = {{Reinforcement Learning Enables Scaling Generalist Robot Policy Improvement}},
author = {Michal Korniak and Marco Bagatella and Ralf R\"omer and Michal Nauman and Yarden As and Pranav Vignesh and Manan Agarwal and Leander Diaz-Bone and others},
year = {2026},
eprint = {2611.06905},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.06905},
}
Juncheng Ma, Jiaxin Li, Jiankai Tu, Bo Liang et al.
Abstract摘要
Egocentric human video offers a scalable source of manipulation priors for robot learning, yet translating raw data scale into generalizable robot policies demands rigorous data curation and training recipes. We introduce Open-HumanScale, the first fully open recipe for scaling egocentric pretraining to 50,000 hours, anchored by Open-HumanScale-50K---a curated egocentric video corpus densely annotated with hand motion and natural language. Leveraging this corpus, we systematically investigate pure human pretraining via closed-loop simulation evaluations and uncover an empirical log-linear …
@misc{ma2026open,
title = {{Open-HumanScale: 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 others},
year = {2026},
eprint = {2611.06817},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2611.06817},
}
Xuhao Qin, Xiyan Huang, Haoyu Dong, Xi Zhang et al.
Abstract摘要
Vision-based tactile sensors capture multiple aspects of physical interaction within a single image stream, including contact appearance, geometry, indentation depth, and tangential deformation. However, existing visuo-tactile policies often directly process tactile observations as images, making it difficult to determine which physical cues contribute to policy learning. This paper presents a controlled real-robot study that compares physically interpretable tactile signals using the LeRobot platform, enhanced with our visuo-tactile perception system. From the same tactile stream, we derive …
Xuemin Hu, Kaixin Li, Qingqing Xia, Yuxin Wang et al.
Abstract摘要
Coordinated whole-body control of humanoid robots remains challenging due to high-dimensional coupled dynamics and substantial physical heterogeneity across joint degrees of freedom (JDoFs). Existing multi-agent reinforcement learning methods either employ coarse agent partitions or directly apply generic heterogeneous-agent algorithms, while hierarchical controllers commonly optimize high-level planning and low-level execution independently, leading to uncoordinated motion. In this paper, we propose H^2MARL, a hierarchical heterogeneous multi-agent reinforcement learning framework with two …
@misc{hu2026h,
title = {{H^2MARL: Hierarchical Heterogeneous Multi-agent Reinforcement Learning for Humanoid Robot Control}},
author = {Xuemin Hu and Kaixin Li and Qingqing Xia and Yuxin Wang and Wei Xie and Bo Tang and Long Chen},
year = {2026},
eprint = {2611.06523},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.06523},
}
Shu'ang Sun, Yibo Qiu, Haoliang Ye, Zan Huang et al.
Abstract摘要
Online reinforcement learning offers an effective approach to precise laboratory manipulation, yet costly real-world demonstrations and limited access to dense process rewards hinder its practical application. We present TwinBridge, a digital twin-driven framework that supplies policy initialization and dense visual rewards for real-world online learning. We construct digital twins by combining 3D Gaussian splatting with physics simulation. Here, simulator states provide task events and geometric progress under physical prerequisites and ordering constraints, yielding rewards for policy …
@misc{sun2026twinbridge,
title = {{TwinBridge: Digital Twin-Driven Policy and Reward Transfer for Real-World Reinforcement Learning in Laboratory Robotics}},
author = {Shu'ang Sun and Yibo Qiu and Haoliang Ye and Zan Huang and Ao Tian and Kan Ding and Ronald Xu and Mingzhai Sun},
year = {2026},
eprint = {2611.06442},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.06442},
}
Robotic manipulation often requires remembering information that is no longer visible, such as an object's location before occlusion or the order of a demonstrated action sequence. A large vision-language planner can read this history, but calling it before every action chunk is slow and costly. We present a three-tier system for the RoboMME benchmark. GPT-6 Astra (System 2) selects a grounded subtask from stored visual history, a $\pi_{0.5}$ vision-language-action policy (System 1) executes it, and a fine-tuned Qwen3-VL-4B monitor (System 1.5) checks after each action chunk whether the …