Social impact of CAVs -- coexistence of machines and humans in the context of route choice

Abstract

Suppose in a stable urban traffic system populated only by human driven vehicles (HDVs), a given proportion (e.g. 10%) is replaced by a fleet of Connected and Autonomous Vehicles (CAVs), which share information and pursue a collective goal. Suppose these vehicles are centrally coordinated and differ from HDVs only by their collective capacities allowing them to make more efficient routing decisions before the travel on a given day begins. Suppose there is a choice between two routes and every day each driver makes a decision which route to take. Human drivers maximize their utility. CAVs might optimize different goals, such as the total travel time of the fleet. We show that in this plausible futuristic setting, the strategy CAVs are allowed to adopt may result in human drivers either benefitting or being systematically disadvantaged and urban networks becoming more or less optimal. Consequently, some regulatory measures might become indispensable.

Figures (8)

• Figure 1: A two-route bottleneck in a city. To reach the other side of the river the drivers have to choose between the alternatives A and B. The everyday choice to minimize travel time can be understood as a repeated game between multiple participants striving to find the option which maximizes a driver's utility.
• Figure 2: The learning and decision processes applied by human drivers (HDVs) and machines (CAVs). HDVs' reasoning is subjective and based on limited access to information. Contrariwise, CAVs have access to complete information on travel times and make optimal collective routing decisions. The interaction between human agents and CAVs may result in any combination of human drivers and CAVs being better off or worse off subject to the strategy applied by CAVs. In particular, the system-wide welfare may improve or deteriorate in the wake of introduction of CAVs.
• Figure 3: Kernel density estimations of mean travel times of HDVs and CAVs for different CAV shares and strategies, based on the final $100$ days of the simulation. In the selfish strategy the CAVs experience shorter while HDVs longer travel times for small CAV shares compared to the situation before CAV introduction. For larger CAV shares both groups' travel times improve, with HDVs gaining more. In the altruistic and social strategies the travel times of CAVs increase and those of HDVs decrease, compared to the travel times before the introduction of CAVs except for the social case with very large shares of CAVs when both groups' travel times decrease. The malicious and disruptive strategies are similar to selfish for small CAV shares however they may cause oscillations and lead to increased travel times of all the vehicles for large CAV shares.
• Figure 4: Comparison of average fractions of CAVs and HDVs on route A for different CAV shares. In the selfish scenario, all CAVs are routed via $A$ for small CAV shares and this fraction decreases for increasing CAV share. The fraction of HDVs on $A$ remains relatively stable. In the social scenario the tendency is exactly opposite, i.e. all CAVs are routed via $B$ (corresponding to fraction $0$ on $A$) for small CAV shares. Altruistic CAVs are all routed via $A$ while HDVs strongly prefer $B$. Malicious CAVs behave similarly to selfish CAVs for small shares, however for large shares their strategy entails routing more vehicles, on average, via A. Finally, the disruptive strategy is, in terms of fractions, on average similar to the malicious strategy.
• Figure 5: Outcomes of replacing a fraction of HDVs by CAVs for different CAV shares and baseline HDV perception bias. CAV advantage$(\tau \slash \rho)$: the ratio of mean HDV travel time averaged over days $301-400$ and mean CAV travel time averaged over days $301-400$. If $\tau \slash \rho>1$ it is better to be CAV than HDV after M-day. Effect of changing to CAV$(\tau_b \slash \rho)$ : the ratio of mean HDV travel time averaged over days $101-200$ and mean CAV travel time averaged over days $301-400$. If $\tau_b \slash \rho>1$, the vehicle which switched from HDV to CAV experiences on average shorter travel times. Effect of remaining HDV$(\tau_b \slash \tau)$: the ratio of mean HDV travel time averaged over days $101-200$ and mean HDV travel time averaged over days $301-400$. If $\tau_b \slash \tau>1$, the vehicle which remained HDV after M-day experiences on average shorter travel times. Perceived effect of remaining HDV$(u_b \slash u)$: the ratio of mean perceived HDV travel time averaged over days $101-200$ and mean perceived HDV travel time averaged over days $301-400$. If $u_b \slash u>1$, the vehicles which remained HDVs after M-day experience on average better perceived travel times.