Joint coordination and routing of autonomous vehicles in road networks with machine learning-based travel time forecasting

This paper addresses the coordination and routing of autonomous vehicles in a traffic network populated only by connected and automated vehicles, by proposing a novel traffic management system (TMS) that jointly determines the optimal vehicle paths in the network, as well as the exact trajectories and speeds on each road and intersection. To this end, two optimization problems are considered and solved within the TMS: a scheduling and motion planning problem, and a routing problem; the latter is formalized in detail in this paper. In addition, a methodology for predicting accurate network travel times based on machine learning is proposed and included in the TMS so that vehicles are routed considering the most accurate information regarding the future expected network state, the exact intersection and road layouts, and the vehicle driving behavior. Different experiments are performed to validate the proposed TMS and provide evidence of its efficacy in reducing congestion and travel times while guaranteeing safety, comfort for passengers, and the feasibility of local vehicle maneuvers such as lane changes and overtaking. The proposed framework contributes to the need of managing the expected growing number of autonomous and connected vehicles within the transport sector, providing a tool to ensure that the expected automation benefits, together with commonly shared sustainable mobility goals, can be actually achieved.