I'm a second-year M.S. ECE student at UCLA. I am fortunate to work with Prof. Anushri Dixit on quadruped locomotion and with Prof. Yuchen Cui on interactive robot learning. I earned my B.S. in Computer Engineering at UCLA, with minors in Mathematics and Bioinformatics, where I worked with Dr. Yizhou Zhao and Prof. Ying Nian Wu. For my senior capstone, I built a pick-and-place robot with Prof. Ankur Mehta.

My goal is to build safe and adaptive robots that learns from interactiving with human and the environment. I'm particularly interested in safe and model based reinforcement learning, online learning, and uncertainty quantification.

Yuanhong Zeng, Anushri Dixit
In Submission
Website  •   Paper  •   Code

TL;DR: Build CVaR-constrained PPO for risk aware RL and use a bandit to adapt the risk level α online from onboard feedback.

Yuanhong Zeng*, Metin Alp Dogan*, Yuchen Cui,
In preparation

TL;DR: When the policy pauses and asks for help when uncertain. Human user draws a trajectory sketch on a tablet, and we guided-diffusion to turn sketch into corrective actions in real time.

Yuanhong Zeng, Yizhou Zhao, Ying Nian Wu,
Conference on Robot Learning (CoRL 2024) CoRoboLearn Workshop
Website  •   Paper  •   Code

TL;DR: Build “digital twins” online and adapt across three spaces—perception, dynamics, and control—to close the sim-to-real gap. This lets us verify plans against real-world observations and correct them before execution.