Robot Learning From Few Demonstrations by Exploiting Data Structure and Geometry

Explore the challenges and advancements in robot learning from limited demonstrations in this 59-minute robotics colloquium. Delve into efficient representations for manipulation skills, optimal control strategies, and intuitive interfaces for meaningful demonstrations. Discover how to leverage the structure and geometry of data, including Riemannian manifolds and tensor factorization techniques, to overcome the constraint of small demonstration datasets. Learn about the integration of learning and control aspects through optimal control frameworks, and how these techniques can be recast as Gauss-Newton optimization problems. Gain insights from Dr. Sylvain Calinon, a Senior Researcher at the Idiap Research Institute, on the latest developments in robot learning, human-robot collaboration, and optimal control.