Teaching a Robot Dog New Tricks: Diverse Quadruped Skills via Combined Reinforcement and Imitation Learning with Adversarial Task Selection

Abstract

This work trains a single quadruped policy to perform diverse skills through reinforcement learning, imitation learning, and adversarial task selection. Specialist teacher policies guide a student, with training focused on its weakest task. The resulting policy performs 22 tasks learned from eight teachers and is deployed on a Unitree B1 robot.

Publication
arXiv preprint arXiv:2610.10601
Date
Links