NomadZ Humanoid Soccer Team

Teaching robots to play soccer using motion imitation reinforcement learning to participate in the RoboCup competition.

Overview

As the motion control team lead of the NomadZ ETH Zurich RoboCup team, I work with a group of Master’s and Bachelor’s students to teach our Booster K1 robots to play soccer using motion imitation reinforcement learning. We generate our motion data from human videos using video reconstruction and motion retargeting, then use AMP (Adversarial Motion Prior) to train our policies. We are currently preparing for the RoboCup 2026 competition.

Training of dribbling policies for our Booster K1 robots using motion imitation reinforcement learning.

Demo

Below are some videos of our current results: