Doctoral Research Seminar on "Memory of Motion: An approach to combine efficiently Model Predictive Control and Neural Networks"


The next Doctoral Research Seminar is on "Memory of Motion: An approach to combine efficiently Model Predictive Control and Neural Networks" by Dr. Olivier Stasse, Senior Research Scientist at CNRS on Nonday, November 8th, 2021 at 11h.

Abstract:

What if we could generate complex movements for arbitrary robots with arms and legs interacting in a dynamic environment in real-time? Such a technology would certainly revolutionize the motion capabilities of robots and unlock a wide range of very concrete industrial and service applications: robots would be able to react in real-time to any change of the environment or unexpected disturbance during locomotion or manipulation tasks. However, the computation of complex movements for robots with arms and legs in multi-contact scenarios in unstructured environments is not realistically amenable to real-time with current computational capabilities and numerical algorithms. The project Memmo aims to solve this problem by 1) relying on massive off-line caching of pre-computed optimal motions that are 2) recovered and adapted online to new situations with real-time tractable model predictive control and where 3) all available sensor modalities are exploited for feedback control going beyond the mere state of the robot for more robust behaviors. We will show the current results we have been able to achieve on the humanoid robot TALOS.

If you are interested to join, please contact the responsible person.