Self-Correcting VLA: Online Action Refinement via Sparse World Imagination

TL;DR

This paper introduces Self-Correcting VLA, a novel approach that enhances vision-language-action models with sparse world imagination and online action refinement, leading to improved robustness and efficiency in robot manipulation tasks.

Contribution

The paper proposes a self-correcting framework for VLA models that integrates sparse world imagination and online action refinement for better physical understanding and self-improvement.

Findings

Achieves state-of-the-art performance on robot manipulation benchmarks.

Reduces steps by 16% while increasing success rate by 9%.

Gains 14% in real-world experiments.

Abstract

Standard vision-language-action (VLA) models rely on fitting statistical data priors, limiting their robust understanding of underlying physical dynamics. Reinforcement learning enhances physical grounding through exploration yet typically relies on external reward signals that remain isolated from the agent's internal states. World action models have emerged as a promising paradigm that integrates imagination and control to enable predictive planning. However, they rely on implicit context modeling, lacking explicit mechanisms for self-improvement. To solve these problems, we propose Self-Correcting VLA (SC-VLA), which achieve self-improvement by intrinsically guiding action refinement through sparse imagination. We first design sparse world imagination by integrating auxiliary predictive heads to forecast current task progress and future trajectory trends, thereby constraining the policy to encode short-term physical evolution. Then we introduce the online action refinement module to reshape progress-dependent dense rewards, adjusting trajectory orientation based on the predicted sparse future states. Evaluations on challenging robot manipulation tasks from simulation benchmarks and real-world settings demonstrate that SC-VLA achieve state-of-the-art performance, yielding the highest task throughput with 16% fewer steps and a 9% higher success rate than the best-performing baselines, alongside a 14% gain in real-world experiments. Code is available at https://github.com/Kisaragi0/SC-VLA.