VLA-Corrector: Lightweight Detect-and-Correct Inference for Adaptive Action Horizon

2026-07-02

AI Summary

Existing vision-language-action (VLA) policies typically employ open-loop execution over fixed action horizons, rendering them vulnerable to minor perturbations in contact-rich tasks and prone to error accumulation and failure. To address this limitation, this work proposes a lightweight detect-and-correct inference framework that leverages a Latent Visual Monitor (LVM) to continuously compare predicted and observed visual feature trajectories in latent space. This comparison dynamically triggers event-driven replanning, while Online Gradient Guidance (OGG) enables closed-loop correction. Notably, the approach integrates seamlessly into diverse VLA models without requiring backbone modifications or retraining. By introducing the first adaptive action horizon mechanism, it significantly enhances robustness in long-horizon contact-intensive tasks while maintaining a low policy invocation frequency.

Abstract

Vision-Language-Action (VLA) foundation models have recently achieved strong progress in embodied intelligence. To reduce policy-call frequency while preserving temporal coherence, most generative policies adopt an action chunk mechanism, executing multiple future actions in an open-loop manner under a fixed action horizon. However, this "predict-then-blindly-execute" paradigm sacrifices closed-loop reactivity: in contact-rich physical interactions, even small local perturbations can rapidly amplify within the open-loop blind spot, leading to compounding errors and ultimately task failure. To address this limitation, we propose VLA-Corrector, a lightweight corrective inference framework for action-chunked VLA policies. Without modifying the backbone policy weights, VLA-Corrector introduces a lightweight Latent-space Vision Monitor (LVM) that continuously compares predicted and actual visual feature evolution, enabling online detection of visual dynamics deviations. Once persistent deviation is detected, the system triggers a truncation event, discards the remaining stale actions, and invokes corrective replanning via Online Gradient Guidance (OGG). The detect-and-correct mechanism of VLA-Corrector naturally induces an event-triggered adaptive action horizon: it preserves long-horizon execution when the current chunk remains reliable, and invokes short-horizon corrective replanning when execution begins to drift. In doing so, VLA-Corrector mitigates the trade-off imposed by static horizons between execution robustness and policy-call frequency. It can be integrated into different VLA models without further retraining the VLA backbone, interrupting compounding errors while preserving much of the efficiency benefit of action chunking and substantially improving robustness in long-horizon, contact-rich robotic manipulation tasks.

Problem

• Vision-Language-Action
• action chunking
• closed-loop reactivity
• compounding errors
• adaptive action horizon

Innovation

• VLA-Corrector
• adaptive action horizon
• latent-space vision monitor
• online gradient guidance
• action chunking