Paper accepted at ICRA 2018
An exciting new paper based on the idea that to use neural network to learn the accessibility sets of controllers.
Nicolas Mansard extended the work by a nice coupling of neural networks, optimization based control and probabilistic roadmaps. This is allowing a huge speed-up of the convergence when solving the optimization problem.
See the nice video made by Mathieu Geisert to illustrate the gain.
Using a Memory of Motion to Efficiently Warm-Start a Nonlinear Predictive Controller
Nicolas Mansard, Andrea Del Prete, Mathieu Geisert, Steve Tonneau, Olivier Stasse,
The PDF of the paper is available here: