Reinforcement Learning for Collaborative Quadrupedal Manipulation of a Payload over Challenging Terrain

Abstract

Motivated towards performing missions in unstructured environments using a group of robots, this paper presents a reinforcement learning-based strategy for multiple quadrupedal robots executing collaborative manipulation tasks. By taking target position, velocity tracking, and height adjustment into account, we demonstrate that the proposed strategy enables four quadrupedal robots manipulating a payload to walk at desired linear and angular velocities, as well as over challenging terrain. The learned policy is robust to variations of payload mass and can be parameterized by different commanded velocities. (Video <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> https://youtu.be/i8kZSYdi9Nk)