arXiv Robotics — research abstracts· Yixuan Yang, Qiang Chen, Run Shao, Shuwei Dong, Wenhao Li, Yunpeng Wang, Wenhan Fang, Haomiao Wu, Xingyu Wang, Lichen Man, Rong Zhang, Xuezhi Cui, Yuan Liu, Bingyan Lu, Jiawei Ma, Xiaoyang Ma, Zhemin Sun, Zhuorui Yao, Yutong Dong, Yuxuan Zhao, Yichong Mu, Jiabao Jiang, Wenxuan Deng, Junling Zhou, Yi Chen, Deyu Zhang, Yiming Yan, Yige Yuan, Shan You, Fei Wang, Haifeng Li, Hongyan Xu, Yichao Cao, Xiu Su·2026-11-05 05:00 UTC· 5 days agoSignalEditorial score75

Compounding Physical RSI: Turning Execution Failure into Inherited Capability

Summary

This research presents Compounding Physical RSI, a self-evolving policy framework that transforms execution failures into heritable capabilities. The method addresses three bottlenecks in physical environments: localizing failure stages, curating corrective experience without costly online rollouts, and safeguarding existing skills via weight-space candidates. Evaluated on the RoboDojo-Sim benchmark across 42 tasks, the proposed model (CSU-AI-07) ranks first with a mean Overall Score of 26.15 and Success Rate of 20.70%, outperforming the next-best entry by 5.92 points and 5.82 percentage points.

Editorial context

The paper introduces Compounding Physical RSI, a framework that converts execution failures into inheritable capabilities through automated gap assessment, behavior-regularized offline reinforcement learning, and skill retention mechanisms, achieving top performance on the RoboDojo-Sim benchmark across 42 tasks.

Source: arXiv Robotics — research abstracts · Read original article ↗

Loading article text…

What the source reports

Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.

Reported numbers

Overall Score
26.15
View original evidence
or the next generation. Evaluated across all 42 tasks of the RoboDojo-Sim benchmark, CSU-AI-07 ranks first on the leaderboard with a three-seed mean Overall Score of 26.15 and Success Rate of 20.70%, outperforming the next-best entry by 5.92 points and 5.82 percentage points.
Open source S6
Reported success rate
20.7%
View original evidence
or the next generation. Evaluated across all 42 tasks of the RoboDojo-Sim benchmark, CSU-AI-07 ranks first on the leaderboard with a three-seed mean Overall Score of 26.15 and Success Rate of 20.70%, outperforming the next-best entry by 5.92 points and 5.82 percentage points.
Open source S6

paper: not_reported

What remains unknown

Not established in the collected evidence: Environment, Control, Data origin.

Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.

Source excerpts and review record

Automatically extracted; no manual editorial approval recorded.

or the next generation. Evaluated across all 42 tasks of the RoboDojo-Sim benchmark, CSU-AI-07 ranks first on the leaderboard with a three-seed mean Overall Score of 26.15 and Success Rate of 20.70%, outperforming the next-best entry by 5.92 points and 5.82 percentage points.
Open source S6

Source：arXiv Robotics — research abstracts · arxiv.org

#Research #Open source #VLA #Robot policies #Reinforcement learning #Sim-to-real