StableVLA: Towards Robust Vision-Language-Action Models without Extra Data
For practitioners deploying VLA models in real-world settings, this work provides a lightweight method to enhance robustness against visual disturbances without requiring additional data or augmentation.
The paper addresses the robustness of Vision-Language-Action (VLA) models under unseen visual disturbances. The proposed Information Bottleneck Adapter (IB-Adapter) improves robustness by an average of 30% over the baseline without extra data, and a 0.5B model achieves competitive robustness with 7B-scale models.
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 both synthetic and physical visual corruptions.