1 paper
Yuhan Xie, Yuping Yan, Yunqi Zhao +2
Despite their strong performance in embodied tasks, recent Vision-Language-Action (VLA) models remain highly fragile under multimodal perturbations, where visual corruption and lin…