$M^2$-VLA: Boosting Vision-Language Models for Generalizable Manipulation via Layer Mixture and Meta-Skills
For robotic manipulation, this work addresses the trade-off between VLM fine-tuning and generalization, offering a method that preserves VLM capabilities while enabling precise control.
M^2-VLA boosts vision-language models for generalizable robotic manipulation by introducing a Mixture of Layers strategy and a Meta Skill Module, achieving effective zero-shot generalization in simulated and real-world environments.
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.