Learning Agile Robotic Locomotion Skills by Imitating Animals

• DOI: 10.15607/rss.2020.xvi.064
• Publication Date: 2020-06-30T14:47:47Z
• Authors: Xue Bin Peng; Erwin Coumans; Tingnan Zhang; Tsang-Wei Lee; Jie Tan; Sergey Levine
• Keywords: FOS: Computer and information sciences; Computer Science - Robotics; Computer Science - Machine Learning; 0209 industrial biotechnology; 02 engineering and technology; Robotics (cs.RO); Machine Learning (cs.LG)

Reproducing the diverse and agile locomotion skills of animals has been a longstanding challenge in robotics. While manually-designed controllers have able to emulate many complex behaviors, building such involves time-consuming difficult development process, often requiring substantial expertise nuances each skill. Reinforcement learning provides an appealing alternative for automating manual effort involved controllers. However, designing objectives that elicit desired behaviors from agent can also require great deal skill-specific expertise. In this work, we present imitation system enables legged robots learn by imitating real-world animals. We show leveraging reference motion data, single learning-based approach is automatically synthesize repertoire robots. By incorporating sample efficient domain adaptation techniques into training our adaptive policies simulation then be quickly adapted deployment. To demonstrate effectiveness system, train 18-DoF quadruped robot perform variety ranging different gaits dynamic hops turns.