Electrical & Computer Engineering Colloquium
November 20, 2019 1:00 P.M.
Refreshments will be served
Reinforcement Learning for Robot Learning
Mitsubishi Electric Research Labs
Abstract: Robot Learning has seen explosive growth and interest in recent years and it is one of the biggest modern challenges for Artificial Intelligence. After a short introduction to general Reinforcement Learning, I will present three robotic systems we worked on and solved using Model-based Reinforcement Learning (MBRL). In particular, I will focus on the modeling techniques we developed, that combine physical and data-driven models using semiparametric Gaussian Process regression, to learn the dynamics of these physical systems. After that, I will introduce a data efficient model-free RL algorithm that we recently developed as an extension of the TRPO algorithm, called Quasi-Newton TRPO. The algorithm is validated in learning the policy of several OpenAI Gym environments. Finally, I will try to point out some future research directions that we are currently interested on.
Bio: Diego Romeres received the M.Sc. degree (summa cum laude) in control engineering and the Ph.D. degree in information engineering from the University of Padua, Padua, Italy, in 2012 and 2017, respectively. Currently he is a Research Scientist at Mitsubishi Electric Research Laboratories, Cambridge, MA, USA. He held visiting research positions at TU Darmstadt, Darmstadt, Germany, and ETH, Zurich, Switzerland. His current research interests include artificial intelligence, machine learning, reinforcement learning, robotics, and system identification theory.