Designing fully autonomous robots presents a pivotal challenge: equipping them with a broad spectrum of motion skills enabling seamless navigation through unstructured environments, dexterous manipulation of objects, and safe interaction with humans. A significant issue lies in the nature of the data employed to learn these motion skills, often characterized by geometric constraints.
Consequently, there arises a necessity to develop learning and control methods that consider the data geometry in order to guarantee the robot's robust and safe performance. Within this context, Leonel Rozo's talk will provide a comprehensive overview of different robot motion learning problems where such geometric constraints may arise. He will discuss how we can solve these challenges through the lens of Riemannian geometry and its integration with probabilistic models, optimal control, and optimization.
Leonel Rozo is Lead Research Scientist at Bosch Center for Artificial Intelligence (BCAI). Learn more here