StableVLA: Towards Robust Vision-Language-Action Models without Extra Data

It is infeasible to encompass all possible disturbances within the training dataset. This raises a critical question regarding the robustness of Vision-Language-Action (VLA) models when encountering unseen real-world visual disturbances, particularly under imperfect visual conditions. In this work, we conduct a systematic study based on recent state-of-the-art VLA models and reveal a significant performance drop when visual disturbances absent from the training data are introduced. To mitigate this issue, we propose a lightweight adapter module grounded in information theory, termed the Information Bottleneck Adapter (IB-Adapter), which selectively filters potential noise from visual inputs. Without requiring any extra data or augmentation strategies, IB-Adapter consistently improves over the baseline by an average of 30%, while adding fewer than 10M parameters, demonstrating notable efficiency and effectiveness. Furthermore, even with a 14x smaller backbone (0.5B parameters) and no pre-training on the Open X-Embodiment dataset, our model StableVLA achieves robustness competitive with 7B-scale state-of-the-art VLAs. With negligible parameter overhead (<10M), our approach maintains accuracy on long-horizon tasks and surpasses OpenPi under real-world visual disturbances. The code will be made publicly available.

Robots that use vision and language to follow instructions often perform well in clean laboratory environments, but they may fail when their visual input is degraded by blur, noise, poor lighting, or physical obstruction. Since real-world environments are unpredictable, it is impractical to collect training data for every possible visual disturbance. This paper studies this problem and shows that current vision-language-action robot models can suffer large performance drops under such conditions. We introduce StableVLA, a lightweight model design that helps robots keep useful visual information while reducing the impact of visual noise. StableVLA does not require extra training data or special robustness training. In both simulated benchmarks and real-world robot experiments, it improves robustness to unseen visual corruptions with only a small increase in model size. These results show that careful model design can help make robot learning systems more reliable for real-world use.