A Gradient-Based Approach for Coordinating Smart Vehicles and Traffic Lights at Intersections

In this letter we consider the coordination of connected and autonomous vehicles (CAVs) with intelligent traffic control signalization to improve the performance of traffic at intersections. We propose a coordination method based on a gradient-based multi-agent system. Specifically, we consider both the traffic light and the approaching vehicles as smart connected agents and we define a suitable edge tension function for each connection between agents. The agents then run a decentralized gradient descent control policy that drives them towards a desirable sequence of intersection crossings. In the literature, the problem of coordinating CAVs at intersections has been widely explored, with different algorithms being proposed. However, the problem of considering both smart traffic lights and smart cars simultaneously is relatively unexplored, especially from a multi-agent system, gradient-based, perspective. The advantages of the edge tension approach lie in its simplicity in terms of computation and in its ability to scale to larger networks. Furthermore, the control framework is applicable under different connected vehicle penetration rates. This flexibility is a desirable advantage given the gradual nature of the implementation of autonomy in real traffic networks. The proposed strategy is evaluated in simulation and significant fuel savings and delay improvements are shown in comparison with fixed-time traffic light control.