PEKING UNIVERSITY · COMPUTER SCIENCE
Learning to perceive,
reason, and act.
I’m an undergraduate in Computer Science and Technology at Peking University, with an expected graduation in July 2027. I plan to pursue a master’s degree at PKU in 2027.
My interests lie in embodied AI and multimodal representation learning. I work on navigation models that learn what to see and anticipate what comes next, and explore video understanding and reliable LLM agents.
WHAT I’M WORKING ON
From learning visual representations to making reliable decisions in the world.
RepresentNav learns a task-driven visual space from complementary vision encoders and predicts future representations before acting, connecting visual perception with latent world modeling.
Turns agent procedures into executable skills with structured actions, lifecycle hooks, and explicit state, making long-horizon workflows more enforceable and token-efficient.
@misc{zhang2026formalskill,
title={Formal Skill: Programmable Runtime Skills for Efficient and Accurate LLM Agents},
author={Xi Zhang and Meijun Gao and Yuntian Zhao and Xinyu Tan and Yilun Yao and Feiyu Wang and Yanshu Wang and Dingsiyi and Tong Yang},
year={2026},
eprint={2605.19604},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2605.19604}
}
Models the subjectivity of video summaries by generating a set of summaries for each video. Combines diffusion-based importance estimation with Bayesian sequential segment selection.
A LITTLE MORE ABOUT ME
Computer Science and Technology · Undergraduate
School of Electronics Engineering and Computer Science
Expected graduation · July 2027
I plan to pursue a master’s degree at PKU in 2027.
LET’S CONNECT
I’m interested in research conversations around embodied AI, multimodal learning, and reliable agents.