Researchers led by Koki Inami have built a system that lets a stock Unitree G1 humanoid race on a running track without help, according to a paper on arXiv. They raced it in three events at the World Humanoid Robot Games 2026: the 100 m, the 400 m and the 1500 m.
The robot follows its lane with the camera in its head, whose images are steadied against the shaking of the body, and a path tracker. Its running policy was trained with reinforcement learning, guided by recordings of human motion. When both feet are off the ground, leg odometry aided by the inertial sensor fills the gap.
Heat in the motors was the main limit. The team penalised hip torque in training, fitted plates to shed heat and capped the speed by motor temperature, with a pace planned for each event.
The G1 finished 13th of 74 in the 100 m and seventh of 35 in the 1500 m. In the 400 m it was 10th of 44 in the first round, then fell in the semifinal after an error in detecting its lane.
Background
Unitree G1
Humanoid robot
The Unitree G1 is a small humanoid robot from China's Unitree Robotics, and research teams, including one at Apple, use it to test new ways of training and controlling humanoids.



