Are autonomous vehicles better off without signals at intersections? A comparative computational study

We model and analyze a futuristic intersection that serves only connected, autonomous and centrally managed vehicles. Under consideration are three control strategies that aim to minimize the total system delay by choosing an optimal trajectory for each vehicle. The first two abandon the concept of signal timing all together whereas the third strategy keeps it. The difference between the two signal-free strategies has to do with a fail-safe buffer requirement introduced to provide redundancy. Each control strategy leads to a unique version of a trajectory-based autonomous intersection management (T-AIM) problem, which is formulated as a mixed integer linear program and solved using a variety of techniques. We found the signal-free strategy holds an overwhelming advantage over the signal-based strategy in terms of efficiency. However, its success is fragile and dependent on the faith in the safety and reliability of the system. When the fail-safe buffer is introduced, the efficiency of the signal-free strategy degrades to a level comparable to that of a properly designed signal-based strategy. Surprisingly, the signal-free strategy with redundancy tends to arrange vehicles in groups that take turns to cross the intersection together. This ``signal-like behavior" manifests itself whenever congestion rises to certain threshold. In addition, solving the T-AIM problem based on signal timing enjoys significant computational benefits, because it eliminates cross conflicts. Thus, the basic logic of signal timing---if not the physical equipment---may survive even after humans are no longer allowed to drive