Particle swarm for path planning in a racing circuit simulation

A racing line is a trajectory that allows car of defined parameters to travel through a given track in minimum time. The growing competition in motorsports racing increases a demand for planning the vehicle path as accurately as possible, what opens a room for computerized methods. In this work, a method of race line generation basing on known optimization methods was presented and implemented using Matlab software. A novel idea is to divide path optimization into several stages - optimizing curved fragments of the track independently, and then joining them by optimizing the transitions at long straight sections. This way the amount of variables being optimized at each time is reduced, what simplifies the problem and the solver is less prone to fall into local minima. The optimization task itself is stated as finding a set of control points positions, such that the interpolated path can be travelled in a minimum time. To answer this problem, three Matlab solvers are implemented and compared: Fmincon (GlobalSearch), Genetic Algorithm and Particle Swarm. The algorithm was tested and presented on realistic racing circuits, that is Silverstone and Circuit De Barcelona - Catalunya. The results are encouraging for further development of a multi-stage approach, however they reveal weaknesses in interpolation approach. The resulting path is limited to the capabilities of restricted control point's positions. Moreover, traditional interpolation methods may not always create a path optimal for a real vehicle.