Human-Centred Foundations for Efficient Collaborative Robotics 

Vincent Bonnet (University of Toulouse / LAAS) from integrated project PI2 will present a webinar on human-centred foundations for efficient collaborative robotics.

A central challenge in assistive and collaborative robotics is to design human-centred controllers that adapt to each operator rather than enforce stereotyped behaviours.

In this talk, Vincent Bonnet will outline three pillars that enable seamless physical interaction. First, accurate, real-time estimation of the human kinodynamic state for safe contact, leveraging recent advances in inertial measurement units and markerless pose estimation. Second, prediction of future human motion and intent in an explainable and safe manner, building on the history of Inverse Optimal Control and Inverse Reinforcement Learning and presenting new extensions that fuse multi-modal signals and encode task constraints. Third, efficient calibration of human and robot models to shrink the sim-to-real gap, via optimal geometric and dynamic identification methods tailored to anthropomorphic systems.

He will discuss perspectives for solving open problems in robustness, uncertainty quantification, and actual deployment.