Development of a NMPC System for Autonomous Vehicles: Integration into the CARLA Simulation Environment and Validation in Complex Scenarios

One of the most groundbreaking innovations of our time is the development of autonomous vehicles (AVs) technology, which has the potential to significantly improve traffic flow, increase road safety, and support environmental sustainability. However, achieving fully autonomous driving (AD) remains a formidable challenge due to the need – among others – for advanced control systems capable of managing highly dynamic, nonlinear, and uncertain driving environments. Indeed, AVs must navigate complex traffic scenarios, make real-time decisions, and satisfy strict safety and performance constraints. In this context, the present thesis focuses on the design and implementation of a control framework based on Nonlinear Model Predictive Control (NMPC) for AD applications. NMPC is particularly well-suited to address the challenges associated with AVs control, as it enables real-time trajectory optimization and complex decision-making, while explicitly accounting for nonlinear vehicle dynamics and input/state constraints. The proposed controller is validated through a comprehensive co-simulation environment that integrates MATLAB/Simulink with the high-fidelity CARLA simulator. This modular and scalable simulation platform simplifies the replication of realistic traffic conditions and the comprehensive evaluation of AD algorithms. Key contributions include the development of essential functionalities within the control system, such as vehicle output data analysis, optimal trajectory generation based on velocity and path curvature, as well as input signal manipulation to govern vehicle behaviour. A dispatching function is introduced to convert the desired acceleration commands produced by the NMPC into actuator inputs, specifically throttle and braking signals, thereby enhancing the practical applicability of the system. This conversion exploits a simplified longitudinal vehicle dynamics model coupled with a simplified gear-shifting model to accurately capture the behaviour of the vehicle under varying operating conditions. The NMPC controller is extensively tuned to balance tracking accuracy, robustness, and computational efficiency, ensuring real-time feasibility. The interface between MATLAB and CARLA is significantly improved to provide a reliable and efficient co-simulation environment. A comprehensive set of challenging driving scenarios is investigated, including 90° turns, overtaking, and suburban junctions with curves within CARLA simulation environment. Additionally, highly complex maneuvers such as extra-urban merging, urban merging with stop conditions, and multi-vehicle roundabout merging are implemented in MATLAB/Simulink, where such scenarios are scarcely found in the literature. Simulation results demonstrate the robustness and adaptability of the proposed NMPC framework across diverse and challenging scenarios. Performance metrics highlight effective trajectory optimization, precise vehicle control, and real-time decision-making capabilities, even in complex traffic interactions. Additionally, the development of modular and reusable Simulink blocks enhances the flexibility and scalability of the control architecture, supporting future research and development in AVs control.