Combining Car-to-Infrastructure Communication and Multi-Agent Reinforcement Learning in Route Choice

Route choice is an important stage in transport planning and modeling. Most of the existing approaches do not consider that road users can nowadays consult new technologies to plan their routes. In this paper, we combine multi-agent reinforcement learning (MARL) and car-to-infrastructure communication (C2I) to deal with route choice. The agents (road users) and the infrastructure interact with each other to exchange traffic information about the road network. The agents send the travel cost of the edges they crossed to the infrastructure. The infrastructure uses these costs to compute shortest paths, which are transmitted to the agents when requested. The agents use such received shortest path to update their knowledge base. The obtained results are compared against a classical MARL approach that does not use C2I communication. Experimental results show that our approach overcomes the compared method in terms of average travel cost.