A program to supervise aggressive drivers

Darya FILATOVA, Associate researcher CHArt Laboratory, France

Human-system interactions are decision control tasks which can be solved by methods of artificial intelligence. Artificial neural networks when combined with reinforcement learning have shown very promising results for control for autonomous vehicles driving in a simulated environment. An acceleration regulator, which controls the autonomous vehicle, is trained using reinforcement learning and attempts to take advantage of gaps between cars, while following a pre-planned path along its lane. The presented in this work results show that agressive-driving policies can be trained to successfully drive comfortably through a crossing, avoiding collision with other cars and being too passive