Randomized CAV routing beats optimal routing strategies, finds new study
When human drivers fear unpredictability, randomness may optimize traffic.
A new preprint from arXiv (July 2026) explores the emerging market of fleets of Connected Autonomous Vehicles (CAVs) competing for passengers. The authors—Grzegorz Jamróz, Łukasz Gorczyca, and Rafał Kucharski—model a city where some vehicles are human-driven (HDVs) and others are CAVs operated by multiple competing fleets. Their counterintuitive finding: under certain conditions, randomized routing outperforms both system optimum and user equilibrium routing.
The key insight: when human drivers have diverse attitudes towards CAVs (some hate them, some love them), introducing randomness into CAV routing creates unpredictable travel times for HDVs. This makes HDV drivers more cautious, reducing congestion and improving overall market efficiency for CAV fleets. However, unchecked randomness could be antisocial. The authors propose augmenting fleet revenue objectives with mean systemwide travel time to drive competition towards social welfare.
- Randomized CAV routing is more efficient than standard optimal routing when human driver attitudes towards AVs are diverse
- Unpredictable CAV travel times scare human drivers into more cautious behavior, reducing congestion
- Paper suggests adding mean systemwide travel time to fleet revenue objectives to prevent antisocial random strategies
Why It Matters
Could reshape how cities regulate autonomous ride-hailing fleets to balance market competition and traffic efficiency.