Machine learning for robotics, particularly in the context of deep learning and reinforcement learning, has demonstrated remarkable results in recent years. From the viewpoint of reliability and robustness of performance, however, there is a significant gap between improving a robotic skill, i.e. to just get better, vs. actually achieving reliability, which is often expressed in success rates clearly above 99%. From experience in the past, reliability requires to exploit the best of all ingredients that are needed in a robotic skill, e.g., including control, perception (vision, force, tactile, etc.), and machine learning. This talk will touch on various related topics that we have been pursuing in recent years, including interesting challenge domains for robot manipulation, exploiting impedance control for contact rich manipulation, deep learning for various perception tasks, and meta reinforcement learning to learn new manipulation skills in fractions of an hour with close to 100% success rate.

Bio: 

Stefan Schaal is a German-American computer scientist specializing in robotics, machine learning, autonomous systems, and computational neuroscience. Stefan held position at MIT, Georgia Tech, ATR before joining USC as a faculty member. He was also a founding director of the Max-Planck Institute for Intelligent Systems in Tubingen. Since 2018 he is a leader of a robotics research team at Google X. Stefan Schaal's interests focus on autonomous perception-action-learning systems, in particular anthropomorphic robotic systems. He works on topics of machine learning for control, control theory, computational neuroscience for neuromotor control, experimental robotics, reinforcement learning, artificial intelligence, and nonlinear dynamical systems.