arXiv · 机器人与计算机视觉· Qixuan Wu (School of Artificial Intelligence, Beihang University, Beijing, China), Xiaole Yang (School of Software, Beihang University, Beijing, China), Yuguang Yang (School of Electronic and Information Engineering, Beihang University, Beijing, China), Canyu Chen (National College for Excellent Engineers, Beihang University, Beijing, China), Linlin Yang (Communication University of China, Beijing, China), Haixia Pan (School of Software, Beihang University, Beijing, China), Guodong Guo (Lane Department of Computer Science and Electrical Engineering, West Virginia University, Morgantown, USA), Baochang Zhang (School of Artificial Intelligence, Beihang University, Beijing, China)·· 2 天前AI 评分75
FAVR:面向视觉语言动作策略的流感知视觉残差对齐方法
FAVR: Flow-Aware Visual Residual Alignment for Vision-Language-Action Policies
AI 导读
预训练的视觉表征虽富含时序信息,但有效整合到现有视觉语言动作(VLA)策略仍具挑战。本文提出 FAVR,一种轻量级框架,通过线性测试时训练(TTT)维持紧凑的时序视觉状态,并在基于流的动作生成过程中检索动作条件的视觉特征。
来源:arXiv · 机器人与计算机视觉 · arxiv.org