Autonomous Vehicle Control Using Particle Swarm optimization in a Mixed Control Environment

Abstract

This work examines effective ways of controlling autonomous vehicles on the roadway while human-operated vehicles remain in use. Particle Swarm optimization is used to control speed, gap, and braking of autonomous vehicles on a merge lane where human-operated vehicles are simulated using the Krauss car-following model. Experiments performed in a simulated environment tested various vehicle densities, ratios of autonomous versus Krauss-operated vehicles, and scenarios where the type of vehicle merging was adjusted. Metrics collected from the simulation include number of merges, collisions, the average merge lane speed, and the average highway or “nonmerging” speed. Results show that the autonomous vehicles are able to learn vehicle following and merging techniques to keep merges and speeds maximal, while keeping collisions minimal.