arXiv：cs.CL· Peidong Wang, Ying Chang, Zhiming Ma, Xiaocui Yang, Shi Feng, Daqi Liu, Yuzhe Yang, Xu Cao, Chuang Wang, Jian Su, Daling Wang·· 16 小时前
AdaRSI：面向具身智能体的按需递归自我改进框架
AdaRSI: On-Demand Recursive Self-Improvement for Embodied Agents
AI 导读
AdaRSI 是一个面向具身智能体的按需递归自我改进（RSI）框架，先由编码智能体改进技能与执行框架，再通过从保存状态重试失败子步骤来归因故障，仅微调仍不可靠的子步骤。
正文
Authors:Peidong Wang, Ying Chang, Zhiming Ma, Xiaocui Yang, Shi Feng, Daqi Liu, Yuzhe Yang, Xu Cao, Chuang Wang, Jian Su, Daling Wang
Abstract:Pretrained vision-language-action models (VLAs) often fail midway through unseen long-horizon tasks. Recursive self-improvement (RSI) offers a remedy, but embodied RSI methods either refine the harness around a frozen executor and so cannot add a missing motion, or train the executor on persistent failures without asking whether a failure lies in how the executor is called or in what it can do. We ask when a self-improving embodied system should change the weights of its executor, and propose AdaRSI, an on-demand RSI framework for embodied agents. A coding agent first improves its skills and harness, then attributes persistent failures by retrying failing sub-steps from saved states, and fine-tunes only the sub-steps that remain unreliable, with demonstrations it collects itself and experience carried over from earlier tasks. On all 50 target tasks of RoboCasa365, with GR00T N1.5 as the executor and no human demonstrations, AdaRSI improves the VLA on all three suites and raises its success on unseen composite tasks from 4.6% to 13.8%. Compared with fine-tuning every sub-step, it reaches slightly higher success while using only 28% of the GPU time and 22% of the self-collected demonstrations, and fine-tuning without a harness trains more yet gains far less.
| Subjects: | Computation and Language (cs.CL); Robotics (cs.RO) |
| Cite as: | arXiv:2610.20474 [cs.CL] |
| (or arXiv:2610.20474v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.20474 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Peidong Wang [view email]
[v1]
Wed, 14 Oct 2026 16:51:09 UTC (5,293 KB)
来源：arXiv：cs.CL · arxiv.org