KERV: Kinematic-Rectified Speculative Decoding for Embodied VLA Models

返回详情 VLA / Vision-Language-Action 每日论文卡