CheckVLA: Execution-Time Verification with Action-Conditioned World Model for Long-Horizon Mobile Manipulation

2026-07-29

AI Summary

This work addresses error accumulation and insufficient anomaly feedback in long-horizon mobile manipulation under open-loop execution. The authors propose an action-conditioned world model–driven online verification and repair mechanism that leverages a frozen pretrained world model to continuously validate the consistency of observed state evolution. By integrating conformal prediction to calibrate risk-aware thresholds and incorporating delay-aware prefix constraints with an event-triggered keyframe library, the system dynamically determines whether and how to rewrite the action sequence. Evaluated on RoboCasa365, the method improves task success rate by 8.5 percentage points, achieving 36.1%, and attains a timely recall of 77.9% at a 5% false positive rate—significantly outperforming baselines relying solely on observations or employing action perturbations.

Abstract

Vision-language-action (VLA) policies commonly execute long-horizon mobile manipulation through open-loop action chunks, issuing multiple actions without receiving new high-level visual input. A committed chunk therefore implies how observations should evolve, but accidental deviations can violate this expectation while the remaining actions continue to propagate the error: commit-time policy confidence cannot react to a deviation that occurs after dispatch, and observation-only anomaly scores lack an action-conditioned reference for separating expected effects from unexplained changes. We propose CheckVLA, which verifies execution with a separately trained, frozen action-conditioned world model. A conformally calibrated risk threshold bounds the episode-level probability of an unnecessary first intervention and determines when to intervene, its exceedance controls how strongly the rewritten suffix retains the superseded chunk, latency-aware hard prefixing restricts replacement to actions that remain deployable, and an event-driven keyframe bank preserves evidence of prior progress across repairs. On RoboCasa365, under a common training recipe and a matched invocation budget, CheckVLA attains a 36.1% average success rate against 27.6% for periodic replanning (+8.5 points). At a matched 5% episode-level false-alarm target, action conditioning raises timely recall to 77.9%, against 48.6% for an observation-only control and 37.9% for an action-shuffled control. These simulation results support action-conditioned verification as a way to restore feedback during chunked execution while keeping the repair consistent with inference latency.

Problem

• long-horizon mobile manipulation
• open-loop execution
• action-conditioned verification
• execution-time anomaly detection
• vision-language-action policies

Innovation

• action-conditioned world model
• execution-time verification
• conformal calibration
• latency-aware replanning
• long-horizon mobile manipulation