VLA-Corrector: Lightweight Corrective Reasoning Boosts Robot Manipulation Robustness

OmniAI-ZJU · hf · 2026-07-06

A Zhejiang University team proposed VLA-Corrector to address the limitations of action chunking in Vision-Language-Action (VLA) models. By introducing a lightweight latent-space visual monitor, it enables adaptive corrective replanning, thereby enhancing robustness for contact-rich manipulation tasks.

Original post →

More from Embodied

Embodied channel →