The Year in AI Papers
The year/Independent research

Paper 2509.09372

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

Yihao Wang, Pengxiang Ding, Lingxiao Li, Can Cui, Zirui Ge, Xinyang Tong, Wenxuan Song, Han Zhao, Wei Zhao, Pengxu Hou, Siteng Huang, Yifan Tang, et al.

Published
Sep 2025
Research lab
Independent
Citations
129
GitHub
2.3K stars

01 In brief

Summary

VLA-Adapter is a novel paradigm for Vision-Language-Action (VLA) models that reduces reliance on large-scale VLMs and extensive pre-training.

It systematically analyzes which vision-language (VL) conditions are essential for bridging perception to action, finding that middle-layer raw features and deep-layer ActionQuery features are most effective, and that using all-layer features generally outperforms single-layer ones.

The proposed Policy module with Bridge Attention autonomously injects optimal conditions into the action space, using a learnable ratio to modulate raw feature injection.

VLA-Adapter achieves state-of-the-art performance on LIBERO (97.3% average success rate) and CALVIN ABC→D (4.42 average length) benchmarks using only a 0.5B-parameter backbone, without robotic data pre-training.

It also offers fast inference (219.2 Hz throughput) and can be trained in 8 hours on a single consumer-grade GPU.

The method remains effective even when the VLM backbone is frozen, and outperforms baselines like OpenVLA-OFT and SmolVLA in such settings.

Real-world experiments confirm its generalization capabilities.

The work provides key findings for VLA design and lowers the barrier to deploying VLA models.

02 From the paper

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/.