Woodruff School of Mechanical Engineering
Mechanical Engineering Seminar
Control Architectures for Agile Legged Systems
Prof. Donghyun Kim
College of Information and Computer Sciences at the University of Massachusetts Amherst
Friday, March 26, 2021 at 1:00:00 PM Add to Calendar
Professor Ye Zhao
To accomplish the dynamic locomotion of legged robots, we need a systematic understanding of hardware, real-time controls, dynamics, and motion planning. Therefore, we need a control architecture with full consideration of hardware and software to realize human- and animal-level agility in robotic systems but even the current state-of-the-art techniques are far from the goal. In this talk, I will explain challenges in dynamic motion control in terms of classical control techniques (e.g. bandwidth of feedback control, uncertainty, and robustness) and high-level planning (e.g. step planning, visual perception, and trajectory optimization), and present my approaches to overcome the difficulties. I will showcase dynamic locomotion of various legged platforms and explain how the hardware and control formulations are related, which evident why a systematic understanding is critical to accomplish dynamic locomotion control. The tested robots include point-foot bipeds (Hume, Mercury), robots using liquid-cooling viscoelastic actuators (Draco), and a quadruped robot using proprioceptive actuators (Mini-Cheetah). I will also present recent results of the Mini-Cheetah-Vision robot which embeds a perception system to improve walking performance over rough terrains.
Donghyun Kim is an Assistant Professor of College of Information and Computer Sciences at the University of Massachusetts Amherst. His primary research area is in dynamic locomotion of legged systems with a focus on the development of a hierarchical control framework and its experimental validation. During his Ph.D. at the University of Texas at Austin, Donghyun developed frameworks including joint-level feedback control, whole-body control, footstep planner, and its robustness analysis, for passive-ankle biped robots. During his postdoctoral research at MIT, he demonstrated Mini-Cheetah running up to 3.7 m/s with his new integrated control framework including a new whole-body controller and model-predictive control. Currently, he is extending his research area to a perception-based high-level decision algorithm to push forward the robots' athletic intelligence.