RA-L 2026
Learning Whole-Body Humanoid Locomotion via Motion Generation and Motion Tracking
A terrain-aware motion generator adapts human motion priors for coordinated whole-body locomotion on a G1 humanoid, using onboard perception and computation.
Zewei Zhang
I study robot learning, with a focus on legged locomotion and whole-body control.
I completed my MSc in Mechanical Engineering at EPFL, with research experience at EPFL’s BioRob and ETH Zurich’s Robotic Systems Lab.
ANYmal / G1 47 seconds
About me
During my MSc at EPFL, I worked with Prof. Auke Ijspeert at the Biorobotics Laboratory and Prof. Marco Hutter at ETH Zurich’s Robotic Systems Lab.
Before EPFL, I studied at Waseda University in Tokyo, where I focused on space exploration. Outside research, I enjoy basketball and hotpot.
Selected publications
My research explores how motion priors can adapt to real-world terrain through online optimization, residual policies, and terrain-aware motion generation. Research projects ↗
CoRL 2025
Motion Priors Reimagined: Adapting Flat-Terrain Skills for Complex Quadruped Mobility
A high-level residual policy adapts dog motion priors to uneven terrain, stairs, and goal-directed navigation on ANYmal.
IROS 2024 · Oral presentation
Online Optimization of Central Pattern Generators for Quadruped Locomotion
Online Bayesian optimization adapts biologically inspired CPG parameters to changing commands, terrain conditions, and payloads on a Go1 quadruped.
Collaborative work
arXiv 2026 · Under review
Humanoid Mobility via Attention-based Contacts
An attention-based contact planner and contact-conditioned controller for whole-body humanoid mobility.