Robotics & Embodied AI Researcher
I am currently a senior undergraduate student at Tsinghua University, working on robotics and embodied intelligence. I recently spent a wonderful summer at UC Berkeley's EMBER Center, advised by Prof. Shankar Sastry. I have also been fortunate to work closely with Prof. Yiming Li, Prof. Hang Zhao, and Prof. Shuguang Li.
I particularly enjoy bridging learning algorithms and real robotic systems, where physical interaction not only tests our ideas but also inspires new ones.
I am currently seeking PhD opportunities starting in Fall 2027, and I would be excited to connect and explore potential research opportunities and collaborations.
My research has spanned robot design, humanoid locomotion, tactile-informed robot learning, and manipulation. More recently, I have become particularly interested in agentic robot self-improvement.
My research interests include:
Life always comes first! 😏
I can hardly imagine life without sports. I've played basketball 🏀 for eight years, and since coming to Tsinghua, I've also picked up tennis 🎾 and started working out 💪.
I love traveling and spending time in nature 🏞️. One of my biggest dreams is to travel around the world, explore its hidden corners, and discover all the wonders it has to offer 🤩.
Recently, I've also gotten into photography 📸. I'm still very much a beginner, but that's part of the fun—there's always something new to learn and capture! 🚀
* Equal contribution
TactileStep integrates deployable sole-pressure feedback with phase-aware reinforcement learning to regulate foot–terrain contact, enabling softer landings and more stable support during humanoid locomotion.
An end-to-end humanoid parkour framework combines foothold saliency, gated temporal memory, and alternating-step regularization to achieve precise foot placement on sparse terrain using onboard depth sensing.
An end-to-end framework uses reinforcement learning to design object-specific origami grippers from target meshes, translating optimized geometries into manufacturable crease patterns and single-motor tendon-driven hardware for reliable, high-load enveloping grasps.
SimRecon reconstructs object-centric, simulation-ready scenes from real videos through a Perception–Generation–Simulation pipeline designed for high visual fidelity and physical plausibility.
to Present
College of AI, Tsinghua University
Advisor: Prof. Yiming Li
Research: Humanoid Robot Learning & Agentic Robot Self-Improvement
to
University of California, Berkeley
Advisor: Prof. S. Shankar Sastry
Research: Robot Manipulation
to
Department of Electronic Engineering, Tsinghua University
Advisor: Prof. Yueqi Duan
Research: Real-to-Sim Reconstruction
Department of Mechanical Engineering, Tsinghua University
Advisor: Prof. Shuguang Li
Research: Computational Design & Fabrication of Origami Robots