Reinforcement learning in cooperative driving

Abstract

We addressed the problem of jointly selecting communications and vehicular planning strategies with the goal of optimizing the driving utility like the speed of the car. Much related work assumes the communications channels as in nite resource. However, our premise is one should optimize both communications and vehicular planning jointly as communication is not a free resource. We performed various simulations to verify our claim and results show that having communication can increase the driving utility while optimizing both planning and communications. We framed the problem as MDP which is then solved using Reinforcement learning techniques to learn the optimal policy. We further extended our work and build a simulator that is more close to the real-world for future work.