Rewind-IL: Online Failure Detection and State Respawning for Imitation Learning

2026-04-17

AI Summary

This work addresses the unrecoverable failures in long-horizon chunked imitation learning caused by execution drift during deployment by proposing a training-free online safeguarding framework. The approach uniquely integrates zero-shot failure detection based on Temporal Inconsistency-based Drift Estimation (TIDE) with a vision-language model–guided semantic safe-state rollback mechanism. It leverages conformal prediction–calibrated TIDE detectors, a checkpoint feature bank built from a frozen policy encoder, and a state-reset policy restart procedure to achieve fault tolerance without additional training. Experiments demonstrate that the framework substantially enhances policy robustness in both real-world and simulated long-horizon manipulation tasks and successfully generalizes to flow-matching–based chunked action policies, confirming its versatility and practical applicability.

Technology Category

• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Natural Language Processing: Safety and Robustness
• Planning, Routing, and Scheduling: Replanning and Plan Repair

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Imitation learning has enabled robots to acquire complex visuomotor manipulation skills from demonstrations, but deployment failures remain a major obstacle, especially for long-horizon action-chunked policies. Once execution drifts off the demonstration manifold, these policies often continue producing locally plausible actions without recovering from the failure. Existing runtime monitors either require failure data, over-trigger under benign feature drift, or stop at failure detection without providing a recovery mechanism. We present Rewind-IL, a training-free online safeguard framework for generative action-chunked imitation policies. Rewind-IL combines a zero-shot failure detector based on Temporal Inter-chunk Discrepancy Estimate (TIDE), calibrated with split conformal prediction, with a state-respawning mechanism that returns the robot to a semantically verified safe intermediate state. Offline, a vision-language model identifies recovery checkpoints in demonstrations, and the frozen policy encoder is used to construct a compact checkpoint feature database. Online, Rewind-IL monitors self-consistency in overlapping action chunks, tracks similarity to the checkpoint library, and, upon failure, rewinds execution to the latest verified safe state before restarting inference from a clean policy state. Experiments on real-world and simulated long-horizon manipulation tasks, including transfer to flow-matching action-chunked policies, demonstrate that policy-internal consistency coupled with semantically grounded respawning offers a practical route to improved reliability in imitation learning. Supplemental materials are available at https://sjay05.github.io/rewind-il

Problem

• Imitation Learning
• Failure Detection
• State Respawning
• Action-chunked Policies
• Long-horizon Manipulation

Innovation

• Rewind-IL
• failure detection
• state respawning
• action-chunked imitation learning
• temporal inter-chunk discrepancy