Robust adaptive learning control for different classes of dissipative vehicle systems

Abstract

This paper presents a methodology that leverages learning techniques and robust control theory to design an adaptive controller for a wide class of linear dynamical dissipative vehicle systems. In particular, learning techniques such as neural networks are used as adaptive learning blocks in the feedback loop with the system under control to update the controller parameters. In order to guarantee the stability of the closed-loop system, a library of parametrized controller blocks that satisfy either the strictly negative imaginary property (SNI), in the case of the negative imaginary system (NI), or the strictly positive real property (SPR) in the case of a positive real system (PR), is developed. The parameters in these controllers are learned using a chosen learning block. The main advantage of including a learning block is to continuously improve performance in the presence of any uncertainty in the environment and the changes in the system's dynamics. This is achieved by allowing the learning block to update the controller parameters based on a defined cost function. Simulation flights testing a quad-copter system are given to illustrate our approach.