Continuous Versatile Jumping using Learned Action Residuals
Yuxiang Yang¹, Xiangyun Meng¹, Wenhao Yu², Tingnan Zhang², Jie Tan², Byron Boots¹
University of Washington, Google Deepmind
Overview
l4dc_hq.mp4Continuous, versatile jumping on the real robot
Forward
Backward
Sideways
Turning
Additional Results in Simulation
Effect of Residual Policy
Residual policy stabilizes the robot while in air.