The Tradeoff Between Altruism and Anarchy in Transportation Networks

In this paper, we ask how a system designer should endow autonomous vehicles with general routing policies that are guaranteed to prove beneficial in a wide variety of networks and adoption rates. Previous work has found that programming autonomous vehicles to be altruistic (i.e., choosing routes in consideration of their impact on aggregate congestion) can guarantee improvements in traffic congestion, provided that enough of the vehicles are autonomous. On the other hand, it is known that if not all vehicles are autonomous, altruistic autonomous vehicles can actually cause significant increases in traffic congestion. Moreover, the benefits of altruistic au-tonomous vehicles depend in complex ways on the fraction of vehicles that are autonomous, complicating the designer's decision. In this paper, we derive the optimal altruism levels for autonomous vehicles which obtain significant benefits while limiting the perverse effects of partial adoption, all without requiring the designer to know the fraction of vehicles that are autonomous. We demonstrate that our proposed altruism levels ensure significant improvements in routing efficiency with respect to previously-known worst-case guarantees.