Hi, I’m Bowen Yang (杨博文)

I am a junior undergraduate at Shanghai Jiao Tong University, Paris Elite Institute of Technology, majoring in Information Engineering and French. I have the privilege of working with Prof. Junchi Yan and Prof. Xiaosong Jia.

RoboMaster competition — on-robot debugging and systems check (© RoboMaster).

I am set to graduate in June 2027 and am actively seeking research internship opportunities in robotics, embodied AI, and robot learning.

Research

My research focuses on robotics and vision–language–action (VLA) models, with an emphasis on how multimodal representations affect perception, action generation, and generalization in embodied systems.

At RethinkLab, Shanghai Jiao Tong University, I work with Prof. Junchi Yan and Prof. Xiaosong Jia on robust VLA policy learning, representation alignment, and mechanism interpretability. My recent work studies how to guide action-token attention toward task-relevant factors such as object grounding, temporal skill recognition, and geometric perception.

My broader technical interests include SLAM, motion planning, sensor fusion, embedded systems, and robotic autonomy, shaped by extensive hands-on experience from RoboMaster competitions, where I served as Vice-Captain and Algorithm Lead and earned two consecutive national championships.

Back-to-Back National Championships

RoboMaster 2024 National Champion — Team Jiao Loong — **National Champion 2024**

RoboMaster 2025 National Champion — Team Jiao Loong — **National Champion 2025**

Featured Project

GuidedVLA overview: action attention specialization for object grounding, skill logic, and spatial geometry

RSS 2026 · Open Source

GuidedVLA: Specifying Task-Relevant Factors via Plug-and-Play Action Attention Specialization

Xiaosong Jia^*, Bowen Yang^*, Zuhao Ge^*, Xian Nie^*, Yuchen Zhou^*, Cunxin Fan^*, et al. Equal contribution.

We guide VLA action decoders to focus on task-relevant factors by assigning dedicated attention heads to object grounding, temporal skill logic, and spatial geometry. The project now includes the paper, arXiv preprint, code, checkpoints, and dataset.

Interpretable specialization: object, skill, and depth heads expose clearer task factors.
Robust generalization: improved performance across simulation benchmarks and real-robot settings.
Released resources: open-source implementation, model checkpoint, and LIBERO dataset.

Project Paper arXiv Code Checkpoint Dataset

Publications

GuidedVLA: Specifying Task-Relevant Factors via Plug-and-Play Action Attention Specialization

Robotics: Science and Systems (RSS), 2026

Xiaosong Jia^*, Bowen Yang^*, Zuhao Ge^*, Xian Nie^*, Yuchen Zhou^*, Cunxin Fan^*, et al. Equal contribution.

GuidedVLA specializes VLA action-decoder attention heads toward object grounding, temporal skill logic, and spatial geometry.

Publication Project arXiv Code

News

May 2026 – GuidedVLA is now open-source, with code, checkpoint, and dataset released alongside the arXiv preprint.
Apr 2026 – Our paper GuidedVLA: Specifying Task-Relevant Factors via Plug-and-Play Action Attention Specialization was accepted to Robotics: Science and Systems (RSS) 2026.
Oct 2025 – Awarded the RM Award (Algorithm); contributed to SLAM and navigation supporting our team’s National Championship at RoboMaster 2025.
Sep 2025 – Received the National Scholarship 2025 (top 0.2% nationwide).
Sep 2025 – Joined RethinkLab, SJTU as a research assistant.
Aug 2025 – Our team won the RoboMaster University Championship National Champion (2025).
Feb 2025 – Selected as Finalist (Top 2%) in the Mathematical Contest in Modeling (MCM), Problem B.
Sep 2024 – Received the National Scholarship 2024 (top 0.2% nationwide).
Aug 2024 – Aug 2025 – Served as Vice-Captain and Algorithm Lead of Team Jiao Loong, directing work on SLAM, motion planning, and embedded autonomy.
Aug 2024 – Our team won the RoboMaster University Championship National Champion (2024).
Oct 2023 – Joined Team Jiao Loong, RoboMaster Team of Shanghai Jiao Tong University.