Failure feedback does not always improve retry success in robot tasks
Rethinking Self-Repair in Code-as-Policy Agents: What Drives Retry Success?
Robotics
Summary
Sometimes robots try to fix their mistakes by retrying the same task using feedback about what went wrong. This paper shows that just retrying with new attempts can be almost as successful as using specific failure feedback. Giving the robot general guidance or feedback from other tries works about the same as using feedback tied to the exact failure. In some cases, specific feedback might even hurt performance if it conflicts with the task. The authors suggest testing different feedback and retry strategies before trusting failure-specific feedback.
What this means in practice
- •For robotics developers: Improve robot task retry strategies by balancing feedback types and resampling to boost success rates on manipulation tasks.
- •For automation engineers: Design self-correcting code agents that use simple, fixed guidance instead of complex failure-specific feedback to recover tasks reliably.
Authors
Siddharth Aphale, Ayushman Singh
Abstract
Code-as-policy agents often use execution feedback to retry failed programs. Additional sampling and task guidance can also improve success, leaving the contribution of failure-specific feedback unclear. We study this question in simulated robot manipulation. A fixed lookup selects feedback templates using failure labels and the API variant. A frozen code model generates a fresh program after each scene reset. On the same 252 initial failures, Qwen2.5-Coder-7B recovers 85.3% within five retries, compared with 68.3% for resampling with the same retry cap. The advantage persists under fixed total-token budgets. Feedback from other episodes and fixed task guidance yield similar recovery point estimates, with no detected benefit from matching the coarse failure label. In a separate study, a fixed guided prompt used from the first attempt achieves an observed overall success rate similar to that of template-based feedback. Template-based feedback also lowers recovery on a nut-insertion task requiring a rim grasp, consistent with an incompatible center-grasp instruction. Evaluations of robot self-repair should include resampling, mismatched feedback, and concise, task-compatible static guidance as baselines.