$M^2$-VLA: Boosting Vision-Language Models for Generalizable Manipulation via Layer Mixture and Meta-Skills

TL;DR

The paper introduces $M^2$-VLA, a novel approach that enhances vision-language models for robotic manipulation by using layer mixture and meta-skills, improving generalization and efficiency.

Contribution

It proposes the Mixture of Layers strategy and Meta Skill Module to bridge semantic understanding and robotic control, enabling zero-shot generalization in manipulation tasks.

Findings

Effective in simulated and real-world environments

Demonstrates zero-shot generalization capabilities

Key components validated through ablation studies

Abstract

Current Vision-Language-Action (VLA) models predominantly rely on end-to-end fine-tuning. While effective, this paradigm compromises the inherent generalization capabilities of Vision-Language Models (VLMs) and incurs catastrophic forgetting. To address these limitations, we propose $M^2$-VLA, which demonstrates that a generalized VLM is able to serve as a powerful backbone for robotic manipulation directly. However, it remains a key challenge to bridge the gap between the high-level semantic understanding of VLMs and the precise requirements of robotic control. To overcome this, we introduce the Mixture of Layers (MoL) strategy that selectively extracts task-critical information from dense semantic features. Furthermore, to facilitate efficient trajectory learning under constrained model capacity, we propose a Meta Skill Module (MSM) that integrates strong inductive biases. Extensive experiments in both simulated and real-world environments demonstrate the effectiveness of our approach. Furthermore, generalization and ablation studies validate the architecture's zero-shot capabilities and confirm the contribution of each key component. Our code and pre-trained models will be made publicly available.