Embedded Files
Our invited speakers will focus on diverse topics, contributing to understanding the application range of generative physical AI for robotics, as well as scientific and technical challenges from different perspectives, including academia and industry. Applications span from autonomous driving, humanoid robotics, industrial robotics, healthcare, etc.
Christopher Agia
Christopher Agia
Christopher Agia is a Ph.D. candidate in Computer Science at Stanford University, advised by Prof. Jeannette Bohg and Prof. Marco Pavone. His work focuses broadly on reliability, interpretability, and composability of policies toward advancing general-purpose manipulation capabilities. He is a member of the Stanford Artificial Intelligence Laboratory and the Center for Research on Foundation Models.
In robot imitation learning, policies are trained to match the behavior distribution of demonstrations, not to maximize expected return. Thus, shaping the training distribution through curation offers a principled means to improve test-time performance. In this talk, we will discuss recent trends in robot data curation and introduce a problem formulation that explicitly links training data selection to downstream policy performance. We will then present CUPID, a data curation algorithm that combines influence functions with policy gradient techniques to estimate the performance contribution of individual demonstrations and identify those most critical to closed-loop success. Through case studies, we will analyze how performance-based curation can (a) improve performance on mixed-quality datasets, (b) recover robust strategies under distribution shift, (c) isolate spurious correlations, and (d) enhance post-training of generalist policies — using only observed policy outcomes, without additional supervision.
Talk “Data-Centric Understanding of Policy Behavior and Performance with Influence Functions.”
Nadia Figueroa
Nadia Figueroa
Nadia Figueroa is the Shalini and Rajeev Misra Presidential Assistant Professor in the Mechanical Engineering and Applied Mechanics (MEAM) Department at the University of Pennsylvania. She holds secondary appointments in the Departments of Computer and Information Science (CIS) and Electrical and Systems Engineering (ESE), and is a core faculty member of the General Robotics, Automation, Sensing & Perception (GRASP) Lab.
Before joining Penn, Dr. Figueroa was a Postdoctoral Associate in the Interactive Robotics Group at the Computer Science and Artificial Intelligence Laboratory (CSAIL) at the Massachusetts Institute of Technology (MIT), where she was advised by Prof. Julie A. Shah. She earned her Ph.D. in Robotics, Control, and Intelligent Systems from the Swiss Federal Institute of Technology in Lausanne (EPFL), under the supervision of Prof. Aude Billard.
Earlier in her career, Dr. Figueroa was a Research Assistant at the Robotics and Mechatronics Institute of the German Aerospace Center (DLR) and at NYU Abu Dhabi. She holds a B.Sc. in Mechatronics from Monterrey Tech and an M.Sc. in Automation and Robotics from TU Dortmund.
Renaud Detry
Renaud Detry
Renaud Detry is an Associate Professor of Robot Learning at KU Leuven in Belgium, where he holds a dual appointment in the Departments of Electrical and Mechanical Engineering (groups PSI and RAM). He is a steering committee member of Leuven.AI, a member of the ELLIS Society, and a technical advisor for OpalAI.
Previously, he led the Perception Systems group at NASA’s Jet Propulsion Laboratory (JPL) in Pasadena, CA, and served as an Assistant Professor at UCLouvain. He is currently a visiting researcher at the University of Liège and KTH Royal Institute of Technology.
His research focuses on robot learning and computer vision, with applications in manufacturing, agriculture, healthcare, and space robotics. At JPL, he was the machine-vision lead for the surface mission of the NASA/ESA Mars Sample Return campaign.
Talk "Fast Multi-task Diffusion Policy Training Using Two-level Mini-batches."
Page updated
Google Sites
Report abuse