2D or 3D: Who Governs Salience in VLA Models? -- Tri-Stage Token Pruning Framework with Modality Salience Awareness

作者：Zihao Zheng, Sicheng Tian, Zhihao Mao, Lingyue Zhang, Chenyue Li, Ziyun Zhang, Hong Gao, Yuchen Huang, Yutong Xu, Guojie Luo, Xiang Chen · 单位：School of Computer Science, Peking University, ZTE Corporation, School of Artificial Intelligence, Beijing Normal University, School of Computer Science, China University of Geosciences (Wuhan), School of Electronics Engineering and Computer Science, Peking University · 会议/期刊：arXiv · 日期：2026-04-10 · 来源：Low-Level Learning-Based Action Modelling / Input Modelling / 2D Vision Language Action Models with Efficiency / Token Pruning