March 9, 2023 Thursday
Abstract:
Controlling quadruped robots to walk on various terrains has always been key research challenge. In recent years, with the development of AI algorithms, the use of deep reinforcement learning to control the robotic motions has achieved great success. However, movements generated based on deep reinforcement learning are often stiff and unrealistic. Based on such observations, we propose to make the quadruped robot imitate the movement of real dogs from the perspective of bionics, and at the same time use reinforcement learning to enhance its adaptability to different terrains. In this talk, I will introduce the whole procedure from real dog data motion capture, imitation learning, terrain adaptive learning, to real-machine deployment, showing the versatile locomotion control capabilities.
Speaker Bio:
Dr. Tingguang LI
Senior Research Scientist, Tencent Robotics X Lab
Tingguang Li is a senior research scientist at Tencent Robotics X Lab. He got his Ph.D. degree from Chinese University of Hong Kong, and his bachelor degree from Nanjing University. His research interests include reinforcement learning, robotics learning, quadruped robot control. He has published around 20 papers at ICRA, IROS and RAL conferences, and filed 18 patents. His work has been nominated the Best Paper Finalist at ROBIO 2017.