Data-Driven Distributed Intersection Management for Connected and Automated Vehicles

This work addresses the problem of autonomous traffic management at an isolated intersection for connected and automated vehicles. We decompose the trajectory of each vehicle into two phases: the provisional phase and the coordinated phase. A vehicle, upon entering the region of interest, initially operates in the provisional phase, during which the vehicle optimizes its trajectory but is constrained from entering the intersection. Periodically, all the vehicles in their provisional phase switch to their coordinated phase, which is obtained by coordinated optimization of the sequence of the vehicles' intersection usage as well as their trajectories. For the coordinated phase, we propose a data driven solution, in which the intersection usage sequence is obtained through a data-driven online classification and the trajectories are computed sequentially. This approach also allows for the incorporation of macro information such as traffic arrival rates into the solution. The overall algorithm is provably safe and can be implemented in a distributed manner. Finally, we compare the proposed algorithm against traditional methods of intersection management and against some existing literature through simulations. Through simulations, we also demonstrate that the computation time per vehicle remains constant for the proposed algorithm over a wide range of traffic arrival rates.