Phantom Menace: Exploring and Enhancing the Robustness of VLA Models against Physical Sensor Attacks

Vision-Language-Action VLA models revolutionize robotic systems by enabling end-to-end perception-to-action pipelines that integrate multiple sensory modalities, such as visual signals processed by cameras and auditory signals captured by microphones. This multi-modality integration allows VLA...