Safely increasing capacity of traffic intersections with mixed autonomous vehicles using delayed self reinforcement

Capacity at traffic intersections can be increased if all autonomous vehicles (AVs) have traffic-light-to-vehicle-communication (TLVC), leading to better inter-vehicle spacing control and reduction of time needed to start and move through an intersection. However, with a mix of autonomous vehicles, some vehicles with TLVC and some without TLVC, the capacity improvements of TLVC are lost and safety can be compromised with potential collisions. Such scenarios can be expected during the transition to fully connected autonomous vehicles (CAVs). The main contribution of this work is to show that the increased capacity of CAVs with TLVC at traffic lights can be recovered safely for mixed AVs network by using delayed self reinforcement (DSR). The advantage of using DSR, where each autonomous vehicles augment its action using delayed versions of past actions, is that it can keep the spacing errors small (i.e., maintain cohesion) even if centralized communication is not available. Therefore, the use of DSR maintains the capacity of traffic intersections even for mixed AVs network, where some vehicles do not have access to the TLVC. Simulation results show the improvement of capacity and safety with the use of DSR when compared to the case without DSR. In particular, DSR improves the traffic capacity by increasing the traffic flow speed by 38% when TLVC is not available. Furthermore, DSR improves safety by reducing the variation in maximum spacing error by 96% under different mixed-connectivity scenarios at the intersection, and thereby, removes the potential for collision seen in the case without DSR.