摘要: |
Intersections are a major source of delay in urban networks, and reservation-based intersection control for autonomous vehicles has great potential to improve intersection throughput. However, despite the high flexibility in reservations, existing control policies are fairly limited. To increase reservation throughput, we adapt two pressure-based policies for reservations in dynamic traffic assignment. The backpressure policy is throughput optimal in communications networks, but communications networks are significantly different from traffic networks. We propose that congestion propagation can be introduced by modeling each cell in the cell transmission model as a link in a communications network. The finite-buffer limitation on the maximum pressure per cell can be overcome by including queue spillback to previous cells and links. However, a counterexample shows that local pressure-based policies such as backpressure cannot be throughput optimal under user equilibrium route choice. Therefore, we also adapt the ??0 policy to reservations. Its adaptation is more straightforward, although dynamic traffic assignment also prevents proving that ??0 is throughput optimal. Nevertheless, results on the downtown Austin network show that both backpressure and ??0 performed significantly better than first-come-first-served, which has been used in most previous work on reservations. Therefore, although backpressure and ??0 cannot be proven to be throughput optimal, they provide a better alternative to existing policies. |