EKT-Router：具身视觉语言模型的专家-证据对齐路由
原标题：EKT-Router: Expert-Evidence Alignment for Embodied Vision-Language Models
AI 摘要
香港中文大学、Knowin.AI 与香港科技大学（广州）等机构的研究者提出 EKT-Router，一个面向具身视觉语言模型的联合路由框架，同时选择能力专家、可选知识与感知工具，并用验证器筛选有效证据。在 6 个具身问答基准上平均得分 61.4，为对比系统最高；在 10 个通用视觉语言基准上平均 74.7。消融显示学习到的路由是主要贡献，知识与感知互补。
正文节选
EKT-Router: Expert–Evidence Alignment for Embodied Vision–Language Models Weilin Ruan1,2, Yingjie Xu2,3, Yangkai Wei2, Bohan Zhou1,2, Shuaibo Li1, Harold Haodong Chen3,4, Kaiwen Zhou2, Yinchuan Li2, Ying-Cong Chen3, James Cheng1,* 1The Chinese University of Hong Kong 2Knowin.AI 3The Hong Kong University of Science and Technology (Guangzhou) 4The Hong Kong University of Science and Technology CONTACT rwlinno@gmail.com * corresponding author Abstract Embodied vision–language models must answer que