Evolutionary optimization of autonomous vehicle tracks

This paper presents a method for the optimization of reference tracks which is used as maps by an autonomous vehicle. Given a track obtained during a manual driving session by a GPS sensor installed in the vehicle, a reduction of the number of points obtained is needed in order to improve the autonomous tracking and control processing times. It is also needed a noise removal in order to avoid the lateral offset error in automatic controlled driving. The optimization is carried out by an evolutionary strategy approach. After an empirical fine tuning of the algorithm parameters, the experiments show that the algorithm is able to provide reference tracks that result in very closed to the tracks manually optimized by an expert, and with a similar autonomous driving performance.