Shaoxuan Li 李少轩
Undergraduate, Computer Science and Technology
Tsinghua University
I am a third-year undergraduate studying Computer Science and Technology at Tsinghua University, with research interests in computer vision, agents, and robotics.
I have been fortunate to work with Prof. Ranjay Krishna at the University of Washington, and with Prof. Zhiyuan Liu and Prof. Hang Su at Tsinghua. My work so far spans perceiving complex long videos (PerceptionComp), agents that improve models by researching their own training data (AutoSFT), and planners that learn from their own execution (PreFTO). Going forward, I aim to bring these vision and agentic techniques to generalist robots, so that they can act intelligently in the real world.
Feel free to reach out at lsx24@mails.tsinghua.edu.cn.
News
- Oct 2026 Awarded the National Scholarship again (top 5% of Tsinghua students).
- Oct 15th 2026 AutoSFT is released on arXiv.
- Jun 18th 2026 PerceptionComp is accepted to ECCV 2026!
- Mar 27th 2026 PerceptionComp is released on arXiv.
- Oct 1st 2025 Awarded the National Scholarship (top 5% of Tsinghua students).
Publications
* denotes equal contribution.
PerceptionComp: A Video Benchmark for Complex Perception-Centric Reasoning
ECCV, 2026
A benchmark for evaluating complex, long-horizon video reasoning in multimodal LLMs.
PreFTO: Reinforcement Fine-Tuning of Diffusion Planners with Local Execution Feedback
Under review at ICLR 2027
An online RL method that improves diffusion-based autonomous driving planners.
AutoSFT: An Agentic System for Automated Data Research in Supervised Fine-Tuning
arXiv, 2026 · Under review at ICLR 2027
An agentic system that improves language models by systematically improving their training data.
Education
- 2024 – Present Tsinghua University · B.E. in Computer Science and Technology · GPA 3.96/4.00, top 3%
Honors & Awards
- 2026 National Scholarship — top 5% of Tsinghua students
- 2025 National Scholarship — top 5% of Tsinghua students