Intervenant : Sylvain Calinon, IDIAP
Lieu : B314 - Bat Ampere (3ème étage)
The recent developments in robot sensors and actuators bring a new human-centric perspective to robotics. The variety of signals to process, the richness of interaction with the users, and the needs to generate natural movements and safe controllers for human-robot collaboration constitute a formidable area of research for machine learning. An attractive approach to the problem of transferring skills to robots is to take inspiration from the way humans learn by employing various forms of imitation and self-refinement strategies. I will first show that robot programming by demonstration encompasses a wide range of imitation strategies. It goes from the mimicking of the demonstrator's actions to higher-level forms of imitation by extracting the intent underlying the actions. I will then discuss the problem of designing compact probabilistic representations of skills, with a focus on dynamical systems, optimal control and stochastic optimization. The combination of these approaches aims at providing the robot with the capability to generalize tasks to new situations, based on demonstrations observed from the perspective of multiple frames of reference. Examples of applications with a compliant humanoid, a continuum flexible robot and a set of gravity-compensated manipulators will be showcased.
Dr Sylvain Calinon is a permanent researcher at the Idiap Research Institute in Switzerland since May 2014. He is also a lecturer at the Ecole Polytechnique Federale de Lausanne (EPFL) and an external collaborator at the Department of Advanced Robotics, Italian Institute of Technology. He holds a PhD from EPFL (2007) awarded by Robotdalen, ABB and EPFL-Press awards. His research interests cover robot learning and human-robot interaction. Webpage: http://calinon.ch