Building explainable and robust neural networks by using Lipschitz constraints and Optimal Transport

Mardi 9 avril de 15h30 à 16h30 , Amphi Costes.

Luca Zaccarian, chercheur au LAAS-CNRS

Abstract:

For a typical coplanar UAV, we address robust trajectory tracking by way of a hybrid control scheme based on two logic modes, one of them consisting of an unconstrained trajectory tracking scheme, which is only effective locally, due to the inevitable saturation constraints, and the second one based on a constrained set-point hovering stabilization scheme, which is endowed with a gradual acceleration/deceleration mechanism for the desired motion, so as to obtain a path following result. We will illustrate the hybrid local/global switching mechanism, discuss the cascade-based stability analysis for both the local and the global configuration, and show some experimental results obtained by the research group of the Aerospace Engineering department of the Politecnico di Milano (Italy).