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

ArXi:2604.16683v1 Announce Type: cross Imitation learning has enabled robots to acquire complex visuomotor manipulation skills from nstrations, but deployment failures remain a major obstacle, especially for long-horizon action-chunked policies. Once execution drifts off the nstration 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.