Robotics Institute Seminar: Nick Rhinehart, “Learning what matters for model-based robotics”

Join in MY580 or online (Zoom)

Learning what matters for model-based robotics

Many important robot applications are difficult to capture precisely with manually designed objectives, and involve dynamics that are too complex for manually designed predictive models. In this talk, I will present recent research on reward learning and model-based methods for robotics. I will focus in particular on advances in structured and efficient world models for simulation and forecasting, including work that extends stable simulation horizons by more than 80x and work that uses unlabeled LiDAR data to achieve top performance on a public benchmark for realistic multi-agent simulation. I will also discuss complementary work on learning objectives when the desired behavior is difficult to specify directly. More broadly, I will argue for robot autonomy methods that combine stronger predictive models with better ways of capturing what should be optimized. These directions point toward robots that can better capture the structure of real tasks, make more effective use of data, and operate more robustly in complex real-world environments.

Speaker bio

Nick Rhinehart is an Assistant Professor at the University of Toronto, where he is appointed to the Institute for Aerospace Science, cross-appointed to the Department of Computer Science, and a faculty member of the Robotics Institute.

He leads the Learning, Embodied Autonomy, and Forecasting (LEAF) lab, which studies learning-based foundations for reliable model-based robot autonomy. The LEAF lab’s recent work focuses on learning transferable dynamics and world models, adapting learned models for efficient real-time planning, and inferring task objectives from sparse human and language-based supervision.

Prof. Rhinehart joined the University of Toronto after holding positions as a Senior Research Scientist at Waymo Research from 2022–2024 and a Postdoctoral Researcher at the UC Berkeley Artificial Intelligence Research laboratory from 2019–2022. He received a Ph.D. in Robotics from Carnegie Mellon University (CMU). His work received the Best Paper Honorable Mention Award at ICCV 2017 and the Best Paper Award at the ICML 2019 Workshop on AI for Autonomous Driving. Prof. Rhinehart holds undergraduate degrees in Engineering and Computer Science from Swarthmore College.