Guaranteed Performances for Learning-Based Control Systems using Robust Control Theory

Concerning guarantees of unmanned systems performances there are significant differences between model-based and learning-based control methods. Model-based control systems are able to provide guarantees on performance specifications, while learning-based control systems through the training process can further improve performances. The goal of this chapter is to propose control design frameworks with performance guarantees on primary (i.e. safety) performances for unmanned systems in which learning-based agents with various structures can be incorporated. These frameworks require the design of a robust Linear Parameter-Varying (LPV) controller and an optimization-based supervisor through an iterative method. The aim of the iteration is to improve the secondary (e.g. comfort, economy) performances of the unmanned control system and thus, the advantages of the learning-based control elements can be taken into consideration. The effectiveness of the design frameworks are illustrated through simulation examples, e.g. cruise control design for autonomous vehicles and the positioning of an arm of a mobile robot.