Unstructured environments and intricate robot maneuvers pose challenges for traditional robot learning models. This talk explores the potential of generative models, particularly continuous normalizing flows, to capture the complexities of desired robot behaviors (e.g., high-dimensionality, multimodal actions, etc). These powerful models are expressive enough to learn complex robot motion skills, including movements on Riemannian manifolds. We will delve into how flow-based models can be leveraged to learn these dynamic skills while introducing strong stability guarantees. This approach enhances controlled generalization capabilities in robot learning frameworks.

Leonel Rozo is currently a lead research scientist in the Bosch Center for Artificial Intelligence (BCAI). He received the B.Sc. degree in mechatronics engineering from the Nueva Granada Military University Bogotá, Colombia, in 2005, the M.Sc. degree in automatic control and robotics, and the Ph.D. degree in robotics from the Polytechnical University of Catalonia, Barcelona, Spain, in 2007 and 2013, respectively. He led the Learning and Interaction Group at the department at the Department of Advanced Robotics, Istituto Italiano di Tecnologia from 2016 to 2018, where he was also a Postdoctoral Researcher from 2013 to 2016. In 2017 he was awarded an individual fellowship from the highly-competitive call Marie Skłodowska-Curie actions for his project proposal DRAPer. His research interests include robot learning from demonstration, machine learning, control, and Riemannian geometry for robotics.

Event on Zoom: