New evidence reframes the “AVs will fix traffic” narrative
A growing body of real-world data is puncturing one of the most durable assumptions in urban mobility: that autonomous vehicles (AVs) will naturally reduce congestion by driving more efficiently than humans. Empirical research published in *Travel Behaviour and Society*, using U.S. city–level data, points in the opposite direction—finding a 6% increase in vehicle miles traveled (VMT) associated with self-driving vehicle deployment.
That headline figure matters because congestion is not linear. Even modest VMT growth can produce outsized delays, especially in dense downtown networks where curb access, signal timing, and bottleneck geometry already operate near capacity. The study’s central mechanism—“deadheading,” or empty AV travel between passenger trips—is not a theoretical artifact. It is an operational reality of fleet-based autonomy. In a striking illustration, Waymo’s San Francisco fleet reportedly spent nearly half of its 2025 operating hours in deadhead mode, meaning a substantial share of road space was consumed without moving a passenger.
The broader implication is sobering for city leaders, investors, and mobility strategists: automation may improve the driving task while worsening the traffic system, unless pricing, data coordination, and fleet incentives evolve as quickly as the vehicles themselves.
—
Deadheading, induced demand, and the hidden geometry of empty miles
The AV congestion story is less about whether a robot car can merge smoothly and more about how fleets behave at scale. Deadheading emerges from several structural features of autonomous ride-hailing and robotaxi operations:
- Repositioning to chase demand: Vehicles relocate from low-demand zones to high-demand zones, often during peak periods when roads are already saturated.
- Parking and depot logistics: AVs may avoid scarce or expensive curb parking by circulating, returning to depots, or staging in remote areas—adding “invisible” miles.
- Door-to-door convenience effects: When friction drops (no driving, no parking, predictable pickup), demand rises. This is classic induced demand, now applied to automated mobility.
The research also highlights a second-order effect with first-order consequences: mode shift away from public transit. If AVs siphon riders from buses and rail—especially for short urban trips—cities can see a double hit:
- More vehicles on the road (higher VMT)
- Fewer passengers per vehicle-mile (lower system efficiency)
This dynamic is particularly acute where transit is already financially fragile. Declining ridership can reduce farebox recovery, forcing service cuts that further push riders toward private mobility—an urban feedback loop that can end in longer travel times, more volatile commutes, and intensified curb congestion around pickup and drop-off zones.
From an infrastructure standpoint, higher VMT also implies accelerated roadway wear and greater maintenance burdens—costs that are rarely priced into private fleet operations but ultimately land on municipal budgets.
—
Why today’s AV stack optimizes trips, not cities
Technologically, the study’s findings underscore a gap between vehicle-level intelligence and network-level optimization. Many routing and dispatch systems are designed to minimize individual wait times and maximize fleet utilization, but they may under-penalize empty repositioning—especially when the business objective is market coverage and short ETAs.
Several constraints compound the issue:
- Algorithmic incentives misaligned with public outcomes: If the platform is rewarded for fast pickups, it may accept deadhead miles as a cost of doing business—while the city absorbs congestion externalities.
- Dense-core sensing and mapping complexity: Urban environments with double-parking, delivery surges, construction, and unpredictable curb behavior can limit the effectiveness of real-time rerouting and staging.
- Data silos between public and private operators: Transit agencies and AV fleets often operate with disconnected demand forecasts, schedules, and curb policies, leading to duplication rather than complementarity.
- Proprietary platform architectures: Closed systems can resist interoperability that would enable cross-platform pooling, coordinated staging, or shared curb management.
The result is a paradox: AVs can be remarkably competent at navigating a lane, yet collectively behave like a high-frequency trading system for curb space, constantly repositioning to capture marginal demand—creating traffic that looks irrational from a citywide perspective but rational from a platform’s competitive lens.
—
The strategic playbook: pricing empty miles, opening data, and rebuilding multimodal advantage
For business and technology stakeholders, the emerging lesson is that AV success in cities will be determined as much by governance and market design as by autonomy performance. The most credible mitigation strategies target the specific sources of excess VMT while preserving the genuine benefits of automation.
Key interventions gaining traction in policy and industry circles include:
- Dynamic congestion pricing and VMT fees tailored to AV operations
– Time-of-day and zone-based charges that explicitly price deadhead miles
– Real-time pricing that reflects network stress, similar in spirit to London or Stockholm models
- Public–private interoperability via data-sharing mandates and open APIs
– Standardized feeds for curb availability, fleet status, and demand forecasting
– Regulatory momentum is building in places like California and through frameworks such as the EU Digital Markets Act, signaling a shift toward enforced coordination
- Incentives for shared-ride AV models
– Credits, preferential curb access, or pricing discounts for higher occupancy
– A direct counterweight to single-occupant automation that inflates VMT
- Digital twin simulations for AV-aware urban planning
– Cities can test scenarios—fleet growth, pricing regimes, curb rules—before congestion becomes politically irreversible
- Mobility hubs and curb-space redesign
– Converting underused parking into multimodal transfer points that integrate AVs with transit, micro-mobility, and last-mile services
For automakers, Tier-1 suppliers, and Mobility-as-a-Service (MaaS) platforms, the competitive frontier is shifting toward software-defined fleet orchestration, not just autonomy. The winners are likely to be those who can prove—quantitatively—that their systems reduce deadheading, support transit, and operate within pricing and data-sharing regimes that cities increasingly view as non-negotiable.
Autonomous vehicles may still reshape urban mobility, but the evidence now suggests the decisive question is no longer whether AVs can drive—it’s whether the ecosystem can govern empty miles, induced demand, and multimodal displacement before the convenience dividend turns into a congestion bill that cities cannot afford.
By
By
By
