Location: https://argonne.zoomgov.com/j/1615761414?pwd=OS95Wjl6eEdkTmpzcHY4K3VzNzc3dz09
Description:
There have been significant advances in the field of robot learning in the past decade. However, many challenges still remain when studying how robot learning can advance interactive agents such as robots that collaborate with humans, and how interactions can enable more effective robot learning. This introduces an opportunity for developing new robot learning algorithms that can help advance the science of interactive autonomy. In this talk, we will discuss a formalism that learns, conventions, i.e., low-dimensional representations sufficient for capturing non-stationary interactions. We demonstrate how we can influence and stabilize these conventions to achieve desirable outcomes in multi-robot coordination. Finally, we will then talk about some of the challenges of learning such representations when interacting with humans, and how we can develop data-efficient techniques that can tap into different sources of data such as play data, suboptimal demonstrations or can actively learn human preferences. We will end the talk with a discussion of applications of these techniques in assistive robotics.
Bio:
Dorsa Sadigh is an assistant professor in Computer Science and Electrical Engineering at Stanford University. Her research interests lie in the intersection of robotics, learning, and control theory. Specifically, she is interested in developing algorithms for safe and adaptive human-robot interaction. Dorsa received her doctoral degree in Electrical Engineering and Computer Sciences (EECS) from UC Berkeley in 2017, and received her bachelor’s degree in EECS from UC Berkeley in 2012. She has been awarded the Sloan Fellowship, NSF CAREER award, ONR Young Investigator award, AFOSR Young Investigator award, IEEE RAS Early Career award, as well as the Google and Amazon Faculty Research Awards.