Autonomous racing simulation solutions: state of the art evaluation and scenario creation

Autonomous racing simulation solutions: state of the art evaluation and scenario creation This thesis addresses the development and validation of autonomous driving algorithms in a racing context, an environment characterized by high speeds, complex vehicle dynamics, and highly variable scenarios, which distinguish it significantly from urban autonomous driving. The main objective of this work was to design and implement a complete pipeline for the simulation, control, and validation of autonomous vehicles on three-dimensional racing tracks, leveraging realistic simulation environments as essential tools for research and development. Firstly, a comprehensive overview of autonomous vehicle simulation environments is provided, highlighting the strengths and limitations of the main available platforms. The selection of the CARLA platform is motivated by technical and design criteria, including simulation realism, flexibility in scenario creation, support for sensor simulation (LiDAR, radar, and cameras), and integration with external tools via Python APIs and ROS 2 middleware. The necessary hardware and software infrastructure to enable near real-time simulations is also described, including libraries, frameworks, and supporting software such as MATLAB/Simulink and FastestLap. Another key component of the thesis is vehicle dynamics modeling. Simplified 3 Degrees of Freedom (3-DoF) models, suitable for planar simulations, and more complex 14 Degrees of Freedom (14-DoF) models, capable of faithfully representing vehicle behavior in racing scenarios, are analyzed. The principles and methodologies for vehicle modeling used in simulation and control are also discussed. The developed case study integrates perception, planning, and control within CARLA. The pipeline includes importing real track geometry into the simulator, implementing surface detection algorithms to reconstruct the mid-lane, trajectory optimization based on vehicle dynamic models, and vehicle control using ROS 2 and Simulink. This approach allows evaluation of the interaction between different modules and the testing of autonomous driving strategies in simulated yet realistic racing scenarios. The results demonstrate the effectiveness of the developed pipeline and confirm that realistic simulators are essential tools for the design, testing, and validation of autonomous driving algorithms in a racing context. Finally, the thesis outlines potential future developments, including the integration of more sophisticated dynamic models, more complex scenarios, and advanced perception and control techniques, paving the way for further research in high-performance autonomous driving.