ARTIFICIAL INTELLIGENCE ON BOARD: ELECTRIC VEHICLE AUTONOMOUS DRIVING

In a period, in which technological development dominated the socio-economic scene, with an aim of evolving and adapting to human needs, the Artificial Intelligence sector could only become the centre of scientific research. When the keyword becomes "comfort", the automation world is highly considered: machine learning, from this point of view, is of great interest for the human assistance process. The focus of the following thesis research is, in particular, to recreate a case study in which the concept of machine autonomy embraces the theme of mobility. The automotive industry, in fact, has concentrated for many years a large part of its resources in the implementation of assisted driving systems and, more recently, autonomous driving. The new vehicles are equipped with an increasing number of optional and advanced security systems. In detail, the project carried out during this research, which took place at the company bylogix srl of Grugliasco, concerns the problem of the Path Following, i.e. the ability of the vehicle to follow a desired trajectory in relation to factors that depend from the intrinsic characteristics of the car and the surrounding environment. The Path Following topic raises many questions related to the ethical, bureaucratic and penal responsibilities of any high-risk situations, as well as the definition of the priority levels to be attributed to the entities involved. Moreover, as the urban reality is varied and dynamic, it is necessary that such systems can predict the most common situations and react very quickly to external stimuli. The present case concerns the implementation of a controller which, given a predetermined trajectory, can process the optimum steering angle for maintaining the path. In particular, the LQR control has been chosen, starting from a reference signal involving the desired values of the lateral position, the lateral velocity, the yaw angle and the yaw rate, to minimize a cost function with the aim of reducing the deviations between the mentioned values and the real ones. The dynamics of the vehicle, which in the real case is a new generation electric model, has been approximated for this purpose using the Bicycle Model. Through the tests performed on the Matlab / Simulink platform, it was possible to compare the results obtained by the different choices of the weight coefficients of the matrices involved. Finally, the same study was revised for the use of the LQI control to highlight any aspects that can be deduced from the comparison between the various results.