The Role of Differentiation in Tolling of Traffic Networks with Mixed Autonomy

With autonomous vehicles now sharing roads with human drivers, the era of mixed autonomy brings new challenges in dealing with congestion. One cause of congestion is when vehicle users choose their routes selfishly to minimize their personal travel delay rather than a global travel delay, and prior works address this phenomenon using tolling to influence routing choices, but do not address the setting of mixed autonomy. Tolls may be differentiated, meaning different users of a road experience different tolls, or they may be anonymous; the latter is desirable to allay concerns of fairness and privacy, as well as logistical challenges. In this work we examine the role of differentiation in traffic networks with mixed autonomy. Specifically, we first establish differentiated tolls which completely eliminate inefficiency due to selfish routing. We then show the fundamental limitations of anonymous tolls in our setting, and we provide anonymous tolls with mild performance guarantees. We show that in parallel networks, an infinitesimal differentiation in tolls is enough to guarantee optimality, and finally we establish a lower bound on the inefficiency of variable marginal cost tolling in the mixed autonomy setting.