Séminaire ayant lieu dans le cadre du Doctorat en informatique cognitive, en collaboration avec le centre de recherche CRIA           

TITRE : Robot Learning from Demonstration

Sylvain CALINON

Jeudi le 11 décembre 2025 à 10h30

Local PK-5115 (Il est possible d'y assister en virtuel en vous inscrivant ici)             

RÉSUMÉ

This talk explores how robots can efficiently acquire complex manipulation skills from minimal human demonstrations, addressing one of the fundamental challenges in modern robotics. I will present approaches that exploit the inherent structure and geometry of demonstration data to enable few-shot learning, moving beyond traditional imitation learning that requires extensive datasets. The discussion will cover representations for manipulation skills that can capture task variations and coordination patterns, optimal control techniques that bridge learning and control, and intuitive interfaces for meaningful human-robot interaction. Key topics include learning on Riemannian manifolds to handle orientation and manipulability constraints, tensor methods for exploiting multidimensional sensorimotor data, and bidirectional interaction strategies that allow robots to actively collect better demonstration data. I will demonstrate applications ranging from industrial manipulation tasks to assistive robotics, showing how robots can adapt learned skills to new situations and perturbations. The talk will address both the theoretical foundations of demonstration-based learning and practical considerations for deploying such systems in real-world scenarios.

BIOGRAPHIE

Sylvain CALINON is Senior Research Scientist at the Idiap Research Institute in Martigny, Switzerland, and Lecturer at the École Polytechnique Fédérale de Lausanne (EPFL). He heads the Robot Learning & Interaction group at Idiap, with expertise in human-robot collaboration, robot learning from demonstration, and model-based optimization. From 2009 to 2014, he was Team Leader at the Department of Advanced Robotics, Italian Institute of Technology (IIT). He holds a PhD from EPFL (2007), awarded the Robotdalen Scientific Award, ABB Award, and EPFL-Press Distinction. His work focuses on human-centered robotics applications where robots acquire new skills from few demonstrations, developing models that exploit data structure and geometry efficiently. He has received Best Paper Awards in Intelligent Service Robotics (2017) and IEEE RO-MAN (2007), and he currently serves as TC Chair on Model-based optimization for robotics for IEEE RAS.

RÉFÉRENCES

Li, Y., Chi, X., Razmjoo, A., & Calinon, S. (2024). Configuration Space Distance Fields for Manipulation Planning. Robotics: Science and Systems (RSS) - Outstanding Paper Award Finalist.

Shetty, S., Lembono, T., Löw, T., & Calinon, S. (2023). Tensor Train for Global Optimization Problems in Robotics. IEEE RAS Best Paper Award.

Jaquier, N., Rozo, L., Calinon, S., & Buerger, M. (2019). Bayesian Optimization Meets Riemannian Manifolds in Robot Learning. Conference on Robot Learning (CoRL) - Best Presentation Award.