Model Predictive Path Integral Control for Car Driving with Autogenerated Cost map Based on Prior Map and Camera Image

Alexander Buyval, Aidar Gabdullin, Konstantin Sozykin, Alexandr Klimchik

Abstract: Model Predictive Path Integral (MPPI) algorithm is an efficient approach to control an autonomous car. However, it requires a cost map. Preparing a cost map in advance is a task that can be completed only for a static closed unchanging environment, but impossible for real city conditions where the environment is dynamic. In this work, we present a method for automatic creation of the cost map based on map data and image from a camera mounted on a self-driving car. The described approach was tested on the Kia Soul car.