Paper Overview
Field: Machine Learning Authors: Zimo Wen, Yijin Chen, Yuxuan Cao, Wendi Chen, Yanwen Zou, Wenye Yu, Fuhang Kuang, Han Xue, Jun Lv, Chuan Wen, Cewu Lu arXiv: 2610.12424
Summary
A generalist robot should not only perform diverse tasks but also improve through experience, turning what it learns during execution into capabilities that later tasks can reuse. Robot agents that act through code can already repair programs from execution feedback, yet organizing this experience around the task structure that gives it meaning remains a central challenge—so that each repair is attributed to the responsible capability, supported by execution evidence, and validated before it is reused.
The authors introduce RoboRSI, a robot self-improvement system built on Top-Down Skill Refinement (TSR). Key ideas:
- Hierarchical skill decomposition: tasks are broken down into compound, atomic, and base skills, each with scoped responsibilities and explicit input-output contracts.
- Attributed repairs: each execution outcome is attributed to the responsible branch, and modifications are confined to that branch.
- Coordinated agent roles: a Manager, Planner, Engineer, and Reviewer collaborate on planning, execution, diagnosis, and validated release of new skills; humans steer the process through goals and corrective feedback.
- Reusable consolidation: stable skill sequences are further integrated into reusable compound skills.
- On a mobile manipulator, RoboRSI developed multi-object household cleanup capability over 104 development iterations.
- In simulation, it achieved the highest success rates on LIBERO, LIBERO-PRO, LIBERO-Plus, and RoboTwin, exceeding the strongest baselines by 2.7 to 11.0 percentage points.
Results
Full Abstract (excerpt)
> A generalist robot should not only perform diverse tasks but also improve through experience, turning what it learns during execution into capabilities that later tasks can reuse. Robot agents that act through code can already repair programs from execution feedback, yet it remains a central challenge to organize this experience around the task structure that gives it meaning, so that each repair is attributed to the responsible capability, supported by execution evidence, and validated before it is reused. We introduce RoboRSI, a robot self-improvement system built on Top-Down Skill Refinement (TSR). TSR decomposes tasks into compound, atomic, and base skills with scoped responsibilities and explicit input-output contracts, attributes each execution outcome to the responsible branch, and ...
*Auto-collected on 2026-10-11.*