Ye Li

Ye Li

Ph.D. Student @ Tsinghua  
Model Efficiency · Embodied AI
Reinforcement Learning
Control Theory for AI Systems

🎓 160+ Citations · h-index 8

RoboStream: Weaving Spatio-Temporal Reasoning with Memory in Vision-Language Models for Robotics

Published in European Conference on Computer Vision 2026 (ECCV 2026), 2026

Recommended citation: Yuzhi Huang, Jie Wu, Weijue Bu, Ziyi Xiong, Gaoyang Jiang, Ye Li, Kangye Ji, Shuzhao Xie, Yue Huang, Chenglei Wu, Jingyan Jiang, Zhi Wang. "RoboStream: Weaving Spatio-Temporal Reasoning with Memory in Vision-Language Models for Robotics." arXiv:2603.12939, 2026. https://arxiv.org/abs/2603.12939

RoboStream: Weaving Spatio-Temporal Reasoning with Memory in Vision-Language Models for Robotics

Y. Huang, J. Wu, W. Bu, Z. Xiong, G. Jiang, Y. Li, K. Ji, S. Xie, Y. Huang, C. Wu, J. Jiang, Z. Wang

📄 Paper  ·  🔗 Project

RoboStream is a training-free framework for long-horizon robotic manipulation that adds persistent spatio-temporal reasoning and memory to VLM planners via Spatio-Temporal Fusion Tokens and a Causal Spatio-Temporal Graph.