Using model-based optimization and movement learning
Chair objectives
How to generate complex movements for arbitrary robots with arms and legs physically interacting in a dynamic environment?
This chair investigates artificial movement and
biomechanics using advanced numerical methods and reinforcement learning (RL), while targeting effective movements on real robot platforms.
We recently demonstrated the premiere whole-body predictive control on the full-scale humanoid robot Talos using a pre-trained memory of motion.
The objective of the chair is to investigate the foundation of polyarticulated movement by
combining advanced numerical methods (optimal control, constrained second-order optimization)and data-driven.
Program Collaborative AI Themes: Neuroscience and AI, Robotics and AI
RLVS ANITI virtual school on reinforcement learning
ANITI-PRAIRIE-MIAI Robotics wokshop (2020)
Publications and references
Carlos Mastalli et al. “Crocoddyl.” IEEE ICRA 2020
Ewen Dantec et al. “WB-MPC with a memory of motion.” IEEE ICRA 2021
Sebastien Kleff et al. “High frequency nonlinear MPC of a manipulator.” IEEE ICRA 2021
Jason Chemin et al. “Learning to steer a locomotion planner.” IEEE ICRA 2021
Mirabel, Joseph, et al. “Integrating Path Planning and Visual Servoing in Manipulation Tasks.” (2020).
Know more
Running a robot: a permanent challenge
Will we one day be able to exclaim "what an elegant step!" "At the sight of a moving humanoid robot? The roboticist Nicolas Mansard strives to perfect the movements of robots using artificial intelligence ... without seeking to reproduce the human walk identically.