Overcoming Dynamics-Blindness: Training-Free Pace-and-Path Correction for VLA Models

TL;DR

This paper introduces a training-free, inference-time correction method for Vision-Language-Action models to handle non-stationary, dynamic scenarios more effectively, improving success rates significantly.

Contribution

It proposes Pace-and-Path Correction, a novel closed-form operator that enhances VLA models' temporal dynamics handling without retraining.

Findings

Outperforms state-of-the-art training-free wrappers and adaptive methods.

Improves success rates by up to 28.8% in dynamic environments.

Demonstrates effectiveness on the MoveBench diagnostic benchmark.

Abstract

Vision-Language-Action (VLA) models achieve remarkable flexibility and generalization beyond classical control paradigms. However, most prevailing VLAs are trained under a single-frame observation paradigm, which leaves them structurally blind to temporal dynamics. Consequently, these models degrade severely in non-stationary scenarios, even when trained or finetuned on dynamic datasets. Existing approaches either require expensive retraining or suffer from latency bottlenecks and poor temporal consistency across action chunks. We propose Pace-and-Path Correction, a training-free, closed-form inference-time operator that wraps any chunked-action VLA. From a single quadratic cost, joint minimization yields a unified solution that decomposes orthogonally into two distinct channels. The pace channel compresses execution along the planned direction, while the path channel applies an orthogonal spatial offset, jointly absorbing the perceived dynamics within the chunk window. We evaluate our approach on a comprehensive diagnostic benchmark MoveBench designed to isolate motion as the sole controlled variable. Empirical results demonstrate that our framework consistently outperforms state-of-the-art training-free wrappers and dynamic-adaptive methods and improves success rates by up to 28.8% and 25.9% in absolute terms over foundational VLA models in dynamic-only and static-dynamic mixed environments, respectively.