Hi! I am Weijue Bu, an incoming graduate student at the Institute of Trustworthy Embodied Artificial Intelligence (TEAI), Fudan University. I am currently a computer science undergraduate at China University of Mining and Technology (2023 – present). From December 2025 to August 2026 I was a research intern at Tsinghua Shenzhen International Graduate School, working on robotic manipulation and vision-language-model-based planning. I am now an embodied AI algorithm intern at Noematrix in Shanghai, having joined on October 26, 2026.

My research interests include agentic robotics, embodied reasoning, spatial intelligence, and long-horizon robotic manipulation.

My current focus is agentic robotics and embodied reasoning — vision-language models that reason about the physical world and drive robots through long-horizon manipulation tasks. Longer term, I want to work on RSI (recursive self-improvement) in Embodied AI.

News

  • Joined Noematrix as an Embodied AI Algorithm Intern.

  • Awarded the National Scholarship (2025–2026 academic year).

  • Admitted to the Institute of Trustworthy Embodied Artificial Intelligence (TEAI), Fudan University, as an incoming graduate student.

  • Paper RoboStream accepted to ECCV 2026.

  • Paper CALO accepted to IEEE Transactions on Cloud Computing.

  • Paper Conscious Gaze accepted to ICME 2026.

  • Won National First Prize in the "Challenge Cup" Competition.

  • Awarded Academic Potential 1st Prize at SJTU Summer School.

Publications

Peer-reviewed Conference

RoboStream: Weaving Spatio-Temporal Reasoning with Memory in Vision-Language Models for Robotics
Yuzhi Huang*, Jie Wu*, Weijue Bu*, Ziyi Xiong, Gaoyang Jiang, Ye Li, Kangye Ji, Shuzhao Xie, Yue Huang, Chenglei Wu, Jingyan Jiang, Zhi Wang
ECCV 2026 Project Page · Code · arXiv
Conscious Gaze: Adaptive Attention Mechanisms for Hallucination Mitigation in Vision-Language Models
Weijue Bu, Guan Yuan, Guixian Zhang
ICME 2026 arXiv

Peer-reviewed Journal

CALO: Code-Aware and Load-Aware Resource Configuration for Serverless Functions via Deep Reinforcement Learning
Donghong Xu, Weijue Bu, Zhouliang Ye, Shilong Wu
IEEE TCC 2026 Code

Preprints & Under Review

ChainVLA: Chaining Vision-Language-Action Queries through a Unified Execution State for Long-Horizon Manipulation
Weijue Bu*, et al.
Under review Project Page
PiN-Mod: Identity-Consistent Personalized Image Generation
Weijue Bu, et al.
Under review

* Equal contribution.

Open Source

OpenYoink

A free, open-source macOS drag-and-drop shelf and Dynamic Island for files, images, text, and links, with Quick Look previews, optional clipboard history, and multi-display support.

SwiftUI · AppKit · macOS · MIT License

Website · Code · Download

Experience

Noematrix · Current

Embodied AI Algorithm Intern 2026.10 – present

Noematrix, Shanghai

Working on embodied AI post-training.

Tsinghua University

Research Intern 2025.12 – 2026.08

Tsinghua Shenzhen International Graduate School

Long-horizon robotic manipulation and vision-language-model-based planning. Built the RLBench / SIMPLER evaluation pipeline and designed long- and short-horizon task suites; co-developed RoboStream (ECCV 2026) and worked on VLA execution-state memory.

Awards & Honors

  • National Scholarship (2025–2026 academic year)

  • Second Prize, 15th "China Software Cup" Collegiate Software Design Contest, National Finals (Higher Education Group)

  • National First Prize, "Challenge Cup" Unveiling Competition

  • 15th Place, Tianchi Alibaba Mobile Recommendation Algorithm Challenge

  • Academic Potential 1st Prize, SJTU Natural Sciences Institute Summer School

  • Second Prize, RAICOM Jiangsu Division Programming Skills Competition

  • Bronze Medal, ICPC China Invitational

  • Second Prize (Provincial), Lanqiao Cup C/C++ Programming

  • Second Prize (Provincial), Jiangsu Higher Mathematics Competition

  • Third Prize (National), GPLT Group Programming Ladder Tournament

Reflections

Three years ago, efficiency was the lens through which I often viewed computer science. Today, I am increasingly interested in how systems learn, reason, and act across unfamiliar situations. Looking ahead, I wonder how experience can become a lasting source of capability—how a system can learn something new without losing what it already knows.

Embodied AI gives this question a physical form. An action changes the world, and the world returns evidence that can refine the next decision. I hope to build systems that can learn from this exchange, allowing experience to shape both what they know and what they do.