A sensitivity-based heuristic for vehicle priority assignment at intersections

This paper presents a sensitivity-based heuristic to address the dynamic priority assignment problem of connected and autonomous vehicle (CAV) and human-driven vehicle (HDV) at traffic intersections. We exploit sensitivity analysis tools to approximatively predict the CAV's performance violation as a function of the HDV states. Such predictions are then used to decide on a crossing order that preserves optimality and feasibility despite the behavior of the HDV. The proposed algorithm is compared with the baseline first-come, first-serve (FCFS) and mixed-integer nonlinear programming (MINLP) approaches. In the closed-loop simulation, we show that the heuristic is computationally much faster than MINLP and able to retain a close-to-optimal solution, which is far better than FCFS.