RESETTLE: Robotic Recovery through Disagreement-Triggered Retrieval and Efficient Corrective Control

2026-10-08

AI Summary

This study addresses the high latency and delayed error correction in robotic manipulation caused by repetitive reasoning and online optimization. To this end, it proposes a model-agnostic recovery framework that innovatively employs persistent divergence in action proposals as a triggering mechanism. Upon activation, the method leverages a V-JEPA encoder for demonstration retrieval and integrates state-servoing priors with visual residual control to achieve single-step correction, thereby eliminating the need for additional planning or online trajectory optimization. Experimental results demonstrate that the proposed framework reduces computational latency by 74%–93% while significantly improving success rates in simulation. Furthermore, real-world task evaluations validate its policy compatibility and overall effectiveness, highlighting its potential for enabling responsive and robust error recovery across diverse manipulation policies.

Abstract

Reliable robotic manipulation requires timely intervention to correct emerging deviations and restore progress after execution errors. However, recovery methods based on repeated vision-language reasoning or iterative online optimization can incur substantial latency, delaying intervention. To address these challenges, we introduce RESETTLE(Robotic rEcovery through diSagrEement-Triggered reTrievaL and Efficient Corrective Control), a model-agnostic framework that provides computationally efficient recovery at the action-execution interface of frozen robot policies. RESETTLE triggers recovery when two action proposals independently sampled under identical conditioning persistently disagree. It retrieves a same-task demonstration reference using an adapted V-JEPA encoder and combines a state-servo prior with a guarded visual residual to execute one corrective action without online trajectory optimization or additional vision-language reasoning, then returns control to the base policy. Across six base policies in simulation, RESETTLE achieves up to 8.70%, 6.28%, and 6.83% absolute success-rate gains on LIBERO-Plus, Meta-World, and RoboCasa Tabletop, respectively, with further improvements on four real-world tasks using two policies. In QwenPI-based comparisons, its monitoring-and-recovery computation latency is 74.04%--93.57% lower than VoLoAgent's monitoring-and-planning latency for grasp and place tool calls. It also raises Harness VLA's LIBERO-Pro Swap success from 42% to 50%, demonstrating compatibility with high-level agentic planning. Code available at: https://github.com/JIA-Lab-research/RESETTLE

Problem

• robotic manipulation
• error recovery
• intervention latency
• vision-language reasoning

Innovation

• Disagreement-Triggered Recovery
• V-JEPA Retrieval
• Corrective Control
• Model-Agnostic Framework
• Low-Latency Intervention