From July 18 to 21, prof. Zhong-Ping Jiang, from Tandon School of Engineering, New York University (more information can be found in his homepage) visited CDSL. Taking lessons from his seminar and further discussions with the members, it was a fruitful time to share the viewpoints and experiences on various topics of researches.
At July 18, though it was the day that he had the flight to Korea, he directly attended our member’s short research presentations, and gave us pieces of advice to improve our results. For example, Jeong Woo made a presentation about his recent result accepted in IEEE CDC 2019, which presents robustness improvement of learning-based controllers via disturbance observers. As the result aims for offline performance recovery with respect to plant uncertainty and disturbance, prof. Jiang suggested that the result be compared with his online-learning algorithm of “RADP,” because they may complement each other.
At July 19, he made a seminar on Learning for Data-Driven Optimal Control, which is an approach to control design for completely unknown linear/nonlinear systems. Based on techniques from reinforcement learning and dynamic programming, the seminar gave us a lesson how to design a learning-based adaptive optimal controllers. The topic of the seminar began with basic definitions and properties of adaptive dynamic programming, but was smoothly continued to specific topics such as global nonlinear/adaptive optimal control and adaptive optimal tracking with disturbance rejection.
Comments are closed.