VLA-Corrector for Reactive Action-Chunked VLA Models

VLA-Corrector is a lightweight framework that introduces closed-loop reactivity to action-chunked Vision-Language-Action models without modifying backbone weights. It mitigates error compounding in contact-rich physical environments by correcting open-loop execution paths.