Keynote Talk
Friday, August 23, 2024 Meeting Room B
The Importance of Policy Representations
Jens KoberAssociate ProfessorTU Delft, Netherlands
Abstract:
Learning motor skills in robotics has applications ranging from agriculture, manufacturing, to household environments and care. Imitation learning is one of the most prominent approaches in this field. Recent advances on large-scale models have enabled generalists, but are not yet suitable for highly specialized, dexterous, or personalized tasks. For those, we want to learn as efficiently as possible, on the one hand, i.e., from as few demonstrations as possible. On the other hand, it is paramount to ensure safety and stability of the learned movements and to enable cooperation with humans. Choosing the right representation of the policy is paramount to achieve these goals - a straightforward application of neural networks is typically neither data efficient nor ensures stability. In this talk I will explore several policy representations developed for efficient learning (also in settings with multiple reference frames), stability, and suitability for human-robot-interaction. All those ideas will be illustrated with real robot experiments.
Bibliography:
Jens Kober is an associate professor at the TU Delft, Netherlands. He worked as a postdoctoral scholar jointly at the CoR-Lab, Bielefeld University, Germany and at the Honda Research Institute Europe, Germany. He graduated in 2012 with a PhD Degree in Engineering from TU Darmstadt and the MPI for Intelligent Systems. For his research he received the annually awarded Georges Giralt PhD Award for the best PhD thesis in robotics in Europe, the 2018 IEEE RAS Early Academic Career Award, the 2022 RSS Early Career Award, and has received an ERC Starting grant. His research interests include motor skill learning, (deep) reinforcement learning, imitation learning, interactive learning, and machine learning for control.