Abstract: This seminar presents our recent work on data-driven approaches to enhance the capabilities of legged robots and industrial manipulators beyond traditional rule-based methods. We begin by showing how reinforcement learning and imitation learning have advanced quadrupedal robots—enabling them to traverse rough terrain, recover from disturbances, and perform agile maneuvers once limited to lab settings. We then turn to industrial manipulators, applying learning-based approaches to industrial tasks to reduce manual engineering effort. These examples illustrate how data-driven approaches can bridge the gap between theory and deployment, enabling robust and adaptable autonomy across domains. The talk concludes with a perspective on combining learning-based components with structured engineering to achieve generality, reliability, and scalability.

Bio: Joonho Lee is the Head of the AI Group at Neuromeka Inc., where he leads research and development of intelligent industrial robots. His research focuses on enabling robust autonomy for robots operating in dynamic and unstructured settings. He received his Ph.D. from the Robotic Systems Lab at ETH Zurich, where he specialized in robust locomotion and autonomous navigation for legged robots in outdoor settings. His work has been published in top venues such as Science Robotics and has been recognized with several honors, including the Willi Studer Prize, ETH Medal, and IEEE Best Paper Awards.

Lunch Provided - First Come, First Served Basis
Host: Sehoon Ha | Assistant Professor, School of Interactive Computing