Set-membership Data-driven control of a Steer-by-Wire system

Advancements in automotive technology are directed towards by-wire systems, that are all those functionalities where mechanical components can be replaced by electronic actuators. Some of the many advantages of this approach are the absence of mechanical limitations, reduced weight and improved safety. The absence of steering shaft and mechanical linkages allows more design flexibility in the whole vehicle architecture; in particular, more freedom in the design of advanced control strategies, such as lane keeping, autonomous parking and active steering. In this thesis, I had the possibility to design and validate a controller for a Steer-by-Wire system intended for an autonomous vehicle for Bylogix S.r.l, a company located in Grugliasco (TO). The objective was to develop a direct data-driven controller using a Set-Membership Identification approach which leverage convex relaxation techniques for polynomial optimization to design the controller. This method avoids the need for explicit system modelling by relying on input-output data and prior assumptions on the bound of the measurement noise. The controller parameters were computed as the Chebyshev center of the feasible controller parameter set, which is the optimal point-wise estimate for the considered problem. This kind of approach gave the opportunity to focus on the actual control problem rather than the identification of the plant, that is often a very time-consuming and expensive process, that can itself lead to secondary issues in the control design phase. However, a first stage of study of the system was necessary to make the correct a-priori assumptions needed for the formulation of the problem. Then, the design problem is formulated and solved using the experimental data acquired on the vehicle under study. The obtained controller is then tested on a vehicle simulator implemented in Matlab-Simulink environment. Future research could explore data-driven design of alternative and more advanced controller structures, such as MPCs or nonlinear neural network-based controllers.