VLA-Adapter: An Effective Paradigm for Tiny-Scale Vision-Language-Action Model

2025-09-11

AI Summary

This work addresses the high computational cost and parameter inefficiency of existing vision-language-action (VLA) models, which typically rely on large-scale pretraining and billion-parameter backbones. To this end, we propose VLA-Adapter, a lightweight, end-to-end trainable framework. Methodologically, it introduces Bridge Attention for dynamic cross-modal alignment, integrates vision-language alignment injection, and employs a conditional policy module—enabling direct perception-to-action mapping without robot-domain pretraining. Notably, VLA-Adapter achieves state-of-the-art performance using only a 0.5B-parameter backbone, trained end-to-end in under eight hours on consumer-grade GPUs. It attains superior results on both simulated and real-world robotic manipulation benchmarks, significantly reducing training cost and deployment complexity while maintaining competitive inference speed and accuracy.

Technology Category

• Computer Vision: Language and Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Vision-Language-Action (VLA) models typically bridge the gap between perceptual and action spaces by pre-training a large-scale Vision-Language Model (VLM) on robotic data. While this approach greatly enhances performance, it also incurs significant training costs. In this paper, we investigate how to effectively bridge vision-language (VL) representations to action (A). We introduce VLA-Adapter, a novel paradigm designed to reduce the reliance of VLA models on large-scale VLMs and extensive pre-training. To this end, we first systematically analyze the effectiveness of various VL conditions and present key findings on which conditions are essential for bridging perception and action spaces. Based on these insights, we propose a lightweight Policy module with Bridge Attention, which autonomously injects the optimal condition into the action space. In this way, our method achieves high performance using only a 0.5B-parameter backbone, without any robotic data pre-training. Extensive experiments on both simulated and real-world robotic benchmarks demonstrate that VLA-Adapter not only achieves state-of-the-art level performance, but also offers the fast inference speed reported to date. Furthermore, thanks to the proposed advanced bridging paradigm, VLA-Adapter enables the training of a powerful VLA model in just 8 hours on a single consumer-grade GPU, greatly lowering the barrier to deploying the VLA model. Project page: https://vla-adapter.github.io/.

Problem

• Reducing training costs of Vision-Language-Action models
• Bridging vision-language representations to action spaces
• Eliminating reliance on large VLMs and pre-training

Innovation

• Lightweight Policy module with Bridge Attention
• No robotic data pre-training required
• Trains on single consumer GPU in 8 hours