Abstract
Traffic congestion has large economic and social costs. The introduction of autonomous vehicles (AVs) can potentially reduce this congestion, both by increasing network throughput and by enabling a social planner to incentivize users of AVs to take longer routes that can alleviate congestion on more direct roads. We formalize the effects of altruistic autonomy on roads shared between human drivers and autonomous vehicles. In this work, we develop a formal model of road congestion on shared roads based on the fundamental diagram of traffic. We consider a network of parallel roads and provide algorithms that compute optimal equilibria that are robust to additional unforeseen demand. We further plan for optimal routings when users have varying degrees of altruism. We find that even with arbitrarily small altruism, total latency can be unboundedly better than without altruism, and that the best selfish equilibrium can be unboundedly better than the worst selfish equilibrium. We validate our theoretical results through microscopic traffic simulations and show average latency decrease of a factor of 4 from worst-case selfish equilibrium to the optimal equilibrium when autonomous vehicles are altruistic.
Introduction
Traffic congestion is a pervasive issue, often exacerbated by the "Price of Anarchy"—the systemic inefficiency that occurs when individual drivers selfishly choose the fastest route for themselves without regard for the collective impact. While traditional solutions like congestion pricing or tolls have been explored, the advent of autonomous vehicles (AVs) introduces a new, programmable lever for traffic management.
This research addresses the "mixed-autonomy" transition period, where AVs share the road with selfish human drivers. Unlike standard routing models that assume latency increases monotonically with flow, real-world traffic exhibits a "congested regime." Past a certain density, vehicle flow actually decreases as latency spikes. The goal is to leverage the coordination potential of AVs to keep road networks out of these high-latency states.
Modeling Mixed-Autonomy Traffic
The core of the proposed framework is a mathematical model based on the Fundamental Diagram of Traffic, which relates vehicle density to flow. A key insight is that AVs can increase road capacity through platooning—maintaining shorter, safer headways than human drivers.
The model accounts for how the "autonomy level" (the proportion of AVs) shifts the critical density point. By increasing this threshold, AVs effectively expand the "bandwidth" of existing infrastructure.
Algorithms for Coordination
The paper introduces two primary strategies for managing mixed traffic:
- Robust Best Nash Equilibrium (RBNE): This algorithm identifies the "best" selfish equilibrium—one that minimizes total delay—while ensuring the system remains resilient to unexpected spikes in traffic demand. The authors provide an $O(N^4)$ algorithm for this task.
- Best-case Altruistic Nash Equilibrium (BANE): This strategy assumes AV users have an "altruism profile," representing their willingness to accept a slightly longer route to benefit the overall network. This is solved via an $O(|K|N^5)$ algorithm. Interestingly, optimal altruistic routing may involve keeping the longest road congested to preserve the rest of the network—a counter-intuitive property unique to the altruistic setting.
The authors demonstrate that even minimal altruism can prevent a network from tipping into a congested state, leading to system-wide improvements that are theoretically unbounded compared to purely selfish systems.
Simulation and Validation
To validate the theoretical findings, the authors employed SUMO (Simulation of Urban Mobility), a high-fidelity microscopic traffic simulator. They modeled a real-world parallel road network in Los Angeles, comparing routes like the 405 freeway with surface streets like Coldwater Canyon and Laurel Canyon.
The simulations confirmed that moving from a "worst-case" selfish equilibrium to an optimal altruistic routing could reduce average latency by a factor of four. Furthermore, the simplified linear model proposed by the authors closely tracked the complex, non-linear behavior of the microscopic simulator, suggesting its robustness for real-world planning.
Conclusion and Future Work
The research highlights that the programmability of AVs offers a powerful tool for traffic management. Even without universal adoption or high levels of altruism, coordinated routing can significantly improve urban mobility.
Future research will extend these models from parallel road networks to complex city grids with intersections. Additionally, the challenges of decentralized coordination among competing AV providers and more nuanced models of human driver psychology remain open areas for investigation.
Summary updated