Talk: Autonomous Development of Scalable Robot Intelligence

LIVE STREAM: https://uwmadison.zoom.us/j/97583720613?pwd=emZTUytWaGFqUzgxekhWSXF1UXZDdz09

Abstract: Developing intelligent robots capable of solving a wide range of tasks in unstructured environments is essential to the advancement of autonomy. In spite of recent advances in robot learning, existing paradigms often require prohibitive manual labor and domain knowledge, resulting in limited scalability. To boost the capability and versatility of robots, we need practical ways to scale up robot learning without extensive human supervision.

In this talk, I will present my research on enabling robots to acquire general-purpose skills through scalable learning with autonomous choices of environments, goals, and tasks. I will describe how to learn robust visuomotor skills that can handle the variety and uncertainty of the real world through the procedural generation of feasible and diverse environments in simulation. I will further present a class of methods that train robots to effectively reuse and repurpose skills learned from prior experiences for novel sequential tasks by proposing reachable subgoals. Finally, I will demonstrate how to discover novel skills in an open-ended manner through adaptive task generation. The acquired skills can be used for solving a variety of challenging tasks such as tool use and sequential manipulation based on raw sensory inputs. I will conclude with future research directions on developing general-purpose robotic systems through the utilization of broad data sources and modalities.

Bio: Kuan Fang is a post-doctoral scholar in the Department of Electrical Engineering and Computer Sciences at UC Berkeley working with Sergey Levine. He received his Ph.D. degree in Electrical Engineering from Stanford University, advised by Fei-Fei Li and Silvio Savarese. His research interests lie at the intersection of robotics, computer vision, and machine learning, with a focus on developing data-driven methods for general-purpose robotic systems. He is a recipient of the Stanford Graduate Fellowship and the Computing Innovation Fellowship.