Model Predictive Contouring Control for Autonomous Racing

The Thesis aims to develop a Model Predictive Contouring Control (MPCC) algorithm for autonomous car racing. The work is conducted within the TUM Autonomous Motorsport Team, a research group from the Technical University of Munich involved in software stack development for autonomous car racing series. The MPCC is an optimal control formulation that balances tracking accuracy with a time-optimality objective for faster driving. For the developed framework, considerable attention is paid to the formulation of the optimal problem cost function, together with a novel state definition. Due to the nature of MPCC, a time-optimality constraint has to be introduced in the cost function definition, with respect to a more classic trajectory-tracking NMPC scheme. However, if a time-based NMPC model is adopted, the controller could opt for considerably faster or-equivalently- shorter trajectories during its re-optimisation phase, causing the car to leave the path originally evaluated by the planning module. So, to precisely track all the considered inputs and constraints, a model conversion from time to the space domain has to be performed. The relevant innovation relies, therefore, on the exploitation of a spatial domain-based model for control applications. The project first aims to develop the Model Predictive Contouring Control algorithm to implement it in the Python prototype control module of the car. Then, its behaviour and possible improvements regarding the operational quantities involved, such as accuracy of the framework, recursive feasibility, and computational time, are evaluated. Finally, the latter algorithm is implemented in the C++ software stack of the autonomous drive architecture.