Quadruped Robot Parkour
March 16, 2026
·
1 min read
Abstract
Train a quadruped robot parkour policy using RL. The end-to-end policy receives depth image and proprioceptive observation as input and outputs desired joint angles.
Demo Video
The Unitree Go2 with trained policy can climb up 0.5m platform and jump across 0.6m wide gap, shown in the below video.
Authors
YasenJia
(he/him)
Robotics Enthusiast
I am currently a graduate student of BIT(Beijing Institute of Technology, China), majoring in Intelligent Robot and System, supervised by Prof.Yan Huang. Before that, I received my bachelor’s degree in Mechatronical Engineering from BIT.