Robot policies improve themselves during task execution without retraining
RoboDoctor: Training-Free Recursive Self-Improvement for Robot Policies
RoboticsComputer Vision and Pattern Recognition
Summary
Robots can often struggle to adjust their plans when things don't go perfectly during a task. This paper presents RoboDoctor, a way for robots to improve their actions in real time based on what actually happens, without needing to be retrained. The system records what decisions were made, what was tried, and how it turned out to help make better choices later. Tests showed that RoboDoctor helped robots adapt and fix their strategies during tasks more effectively.
What this means in practice
- •For robotics engineers: Enable robot control systems to adapt their actions during physical tasks without retraining models.
- •For automation integrators: Provide adaptive feedback mechanisms to improve robot task success rates based on real-time observations.
Authors
Heyang Yu, Haomin Li, Sihang Li, Jixin Yan, Shiyun Peng, Wanqi Zhong, Tianyu Xin, Yuan Yao, Xiaoyu Tian, Yiming Li
Abstract
Generalist robot policies and language-model agents provide complementary capabilities, but completing a task also requires adapting their use to the consequences of physical interaction. We study execution as a locus of adaptation: feedback guides the current action and supplies evidence for revising subsequent decisions under fixed model weights. We introduce RoboDoctor, a training-free framework that connects diagnosis, grounded action, and persistent experience across three revision targets. The harness allocates control between policy and agent; intervention determines when to assist while preserving useful policy behavior; and operation refines physical methods through observed effects. Linked experience records retain the reason for a decision, the attempted method, its outcome, and conditions for reuse. On a curated archive of 54 RoboDojo-Sim task entries, RoboDoctor achieves a five-dimension mean score of 57.10, with a native success rate of 49.44%. Recorded interactions illustrate strategy changes, selective correction, and grasp revision. These findings characterize execution-time adaptation in the evaluated system; controlled trials remain necessary to isolate the contribution of cross-episode experience.