Design of a NMPC System for Automated Driving and Integration into the CARLA Simulation Environment

The development of autonomous vehicles is one of the most significant and promising technological challenges of the modern era, with substantial potential benefits in terms of road safety, traffic efficiency, and environmental sustainability. However, the complexity of the control systems required for fully autonomous driving demands advanced approaches capable of managing dynamic scenarios and variable operating conditions. In this context, simulators play a crucial role in testing and validating these technologies by providing a safe and controlled environment to replicate diverse driving scenarios without real-world risks. This thesis presents the design of a control system for autonomous driving implemented in MATLAB, co-simulated with the CARLA simulator. The project focuses on the implementation of a Nonlinear Model Predictive Control (NMPC) for advanced path tracking and decision-making functions in realistic and dynamic scenarios. The choice of NMPC is motivated by the need to ensure high performance in nonlinear contexts, while also considering strict constraints on vehicle states and inputs. To enable seamless integration of the NMPC controller, this thesis developed several key functionalities within the control system, including vehicle output data analysis, optimal trajectory generation based on velocity and path curvature, and the manipulation of input signals to govern the vehicle’s behavior. Additionally, system identification was employed to examine the longitudinal dynamics of the vehicle under various throttle and braking conditions, enhancing the control system through a dispatching function that converts desired acceleration into appropriate input commands for the vehicle. A detailed analysis of the NMPC controller's tuning parameters was conducted to achieve a balance between tracking accuracy, robustness, and computation time—critical aspects for real-time system implementation. Moreover, the interface between Matlab and CARLA was significantly improved compared to previous studies, ensuring the controller's proper integration into the CARLA environment. The co-simulation between MATLAB and CARLA provides a realistic and modular testing environment, allowing the simulation of different traffic scenarios and variable road conditions. The results demonstrate how the NMPC controller, supported by the newly developed system functions, can effectively adapt to diverse operating conditions, optimizing the vehicle's trajectory and enhancing real-time decision-making capabilities. This work represents a significant step toward integrating advanced predictive control techniques in autonomous driving systems, highlighting both the advantages and challenges of MATLAB-CARLA co-simulation for autonomous driving applications.