Development of a hierarchical architecture for real-time autonomous vehicle control

The autonomous driving is one of the automotive research challenges of the last years, and even if a lot of technologies steps forward have been taken, it remains an open research issue. So, the aim of this PhD thesis was to improve the current level of dynamic control of an autonomous car using commercial hardware and sensor technologies and respecting their functioning constraints. The research focused on the development and implementation of different logic layers inside the autonomous car framework. The first step was to develop an algorithm to localize the vehicle and estimate the car speeds. In this way, the signals required for the correct operation of the control architecture are added to the input signals provided by the sensors. Two of the most adopted technologies were tested. However, the estimation errors made were too high to guarantee the desired level of operation of the control systems. Therefore, based on the characteristics and issues given by the design of these systems, we developed a vehicle speeds estimator consisting of a combination of both. In this way, even in high non-linear dynamic conditions, the errors on the estimation were reduced, improving the car localization and functioning of the control algorithms. Then, an on-line Path Planning was developed able to define the performances that maximize the car speed in a known track. The focuses were: to ensure a real-time trajectory calculation (updating it with the current vehicle dynamics and environmental conditions); and allow computational cost compatible with the correct functioning of the other systems involved. For this reason, it was chosen an optimization algorithm that allows to maintain a linear cost function simplifying the car model and limiting the computational times. However, in order to ensure the most correct representation of vehicle dynamics a more accurate modelling of the car GG-V has been implemented as constraint equations. Once defined the trajectory, a high-level controller was developed to track the dynamic performances provided by the Path Planning. So, was implemented an algorithm that ensures a feed-back control of the Steering Wheel Angle (SWA), accelerator and brake pedal by dividing the later dynamic model from the longitudinal one. In addition, was added a feed-forward control that tracks the lateral and longitudinal acceleration calculated by the planner. Thus, the performances tracking delay and the feed-back control modelling errors were reduced. Finally, to enhance the lateral and longitudinal stability, an Electronic Stability System (ESC) and Anti-lock Brake System (ABS) controls are developed with the aim of improve the current commercial systems performances. About the lateral stability control, a tracking controller was implemented to define the brake input corrections that must be added to the ones established by the Trajectory Tracking to follow reference yaw rate and side slip angle. Instead, to ensure that the wheels don’t lock, a discrete control allows the wheels to follow a target longitudinal speed. In this way, it was found that comparing with the standard ABS the tuning process takes less time and the brake performances are increased, in terms of reduction of the brake distance; and increased stability during full braking manoeuvres. The different layers ware developed independently achieving their own performance improvement and then are integrated together with specific interfaces. Furthermore, two mathematical modelling types were made starting from available experimental data: a sensor characterization to ensure the input signals have their delays, noise and sample time; and a transfer function model of the Brake-By-Wire system developed by Meccanica 42 srl, to ensure actuation outputs delay and constraints. Thus, the real exchange of signals between the architecture developed and the vehicle model is ensured, and the correct operation of the controls once implemented in the car is verified. A real-time static simulator placed at Meccanica 42 srl is used to develop and test the architecture and compare the performances obtained with the ones of a driver. The results showed that: it was possible to obtain an optimisation and tracking of the trajectory that update in real-time taking into account the current dynamic conditions; some good improvements were achieved both with regard to the estimation of the states and the stability of the vehicle; and it was possible to integrate the various layers together guaranteeing satisfactory dynamic performance even if worse than the individual one.