A Self-Correcting Vision-Language-Action Model for Fast and Slow System Manipulation

TL;DR

This paper introduces a self-correcting vision-language-action framework for robotic manipulation that combines fast action prediction with slow reflection and correction, improving robustness and accuracy in complex tasks.

Contribution

The novel SC-VLA framework integrates a fast system for quick pose prediction and a slow system for failure reflection and correction within a unified model.

Findings

Enhanced manipulation accuracy in simulation and real-world tasks.

Effective failure correction and adaptive learning from corrected samples.

Improved robustness on unseen manipulation tasks.

Abstract

Recently, some studies have integrated Multimodal Large Language Models into robotic manipulation, constructing vision-language-action models (VLAs) to interpret multimodal information and predict SE(3) poses. While VLAs have shown promising progress, they may suffer from failures when faced with novel and complex tasks. To emulate human-like reasoning for more robust manipulation, we propose the self-corrected (SC-)VLA framework, which integrates fast system for directly predicting actions and slow system for reflecting on failed actions within a single VLA policy. For the fast system, we incorporate parameter-efficient fine-tuning to equip the model with pose prediction capabilities while preserving the inherent reasoning abilities of MLLMs. For the slow system, we propose a Chain-of-Thought training strategy for failure correction, designed to mimic human reflection after a manipulation failure. Specifically, our model learns to identify the causes of action failures, adaptively seek expert feedback, reflect on the current failure scenario, and iteratively generate corrective actions, step by step. Furthermore, a continuous policy learning method is designed based on successfully corrected samples, enhancing the fast system's adaptability to the current configuration. We compare SC-VLA with the previous SOTA VLA in both simulation and real-world tasks, demonstrating an efficient correction process and improved manipulation accuracy on both seen and unseen tasks.