Robotaxis and Evacuation: An uncomfortable calculation

Author: Bern Grush

Date Published: April 12, 2026

The rapid growth of shared autonomous vehicle fleets is rightly celebrated for what it promises: fewer crashes, lower emissions, reduced congestion, and mobility for those who cannot drive. But the transition away from private car ownership carries a risk that has received almost no systematic policy attention. What happens when an entire community needs to evacuate immediately, all at once, with no vehicle returning for a second load?

This concern is not entirely new. Writing in Planetizen in 2017, urban planners Michael Boswell and William Riggs [1] asked whether autonomous vehicles could enhance emergency evacuation — envisioning fleets dispatched to homes to rescue people who may lack cars, much as many wish had happened during Hurricane Katrina. Their optimism was reasonable at the time, and their instinct to pair AV deployment with emergency management planning was sound. But they were asking the additive question: can AVs effectively help in an evacuation? The question that has since emerged — and that has received almost no policy attention — is the subtractive one: what happens when AV adoption is successful enough that the car-owning majority shrinks? The calculus is not reassuring.

How will robotaxi fleets be sized?

A robotaxi fleet is not designed for emergencies. It is designed for profitability. Operators size their fleets to match typical daily and weekly demand cycles, not for the busiest day of the year — and certainly not for emergency evacuations. Currently, the highest ratio of fleet size to population (San Francisco) is about 1:1,000. In congested cities we foresee that ratio climbing, given growing acceptance, fleet reliability, expanding service areas, falling ride costs, displacement of human-driven ride-hailing and taxi fleets, public-transit ridership attrition, and a long-promised decline in household vehicle ownership.

Over the next decade or so, it appears feasible for some cities to reach ratios of around one 4-6-person robotic passenger vehicle per 100 to 150 residents in a well-served city — and that may be an optimistic upper bound for a city under 250,000. No operator maintains idle vehicles simply as a reserve against unlikely contingencies. Unlike a utility, which faces regulatory requirements to hold reserve capacity, a mature robotaxi fleet will be sized to maximize revenue.

This is a rational business decision. But for any community that comes to rely substantially on robotaxis as a key mode of transportation, this becomes a public vulnerability issue.

The one-trip constraint

The kind of emergency that demands rapid urban evacuation — an advancing wildfire, a failing dam, a chemical release — has a defining characteristic: there is no second trip. The distance to safety, the speed of the threat, and the condition of the roads all but guarantee that a ground vehicle, once loaded and dispatched, will not be able to return for another family. Every vehicle gets one pass. That is the constraint that makes fleet size so consequential.

To see this, consider a town of roughly 100,000 people — large enough to be genuinely urban, small enough to plausibly face a perimeter wildfire or flood threat. The chart below illustrates four scenarios representing different stages of robotaxi adoption:

Scenario A — Early adoption (roughly 5% car-free households, ~200 robotaxis). The 95% of households with private vehicles load up and drive out. The robotaxi fleet, packed at emergency occupancy of four persons per vehicle, moves 800 people — less than 1% of the population — in its single pass. Most of the car-free 5% must hope for a seat in a neighbour’s car, an emergency-impressed transit bus, or an emergency services vehicle. The number left stranded is small, about 4%, so it can be plausibly addressed by emergency management.

Scenario B — Modest adoption (20% car-free households, ~400 robotaxis). Private vehicles carry roughly 80% of the population. 400 vehicles x 4 passengers, evacuates 1,600 people — less than 2% of the population. 18% of residents have no private car or robotaxi evacuation path. In a town of 100,000, that’s 18,000 people depending on neighbours, improvised carpooling, emergency vehicles, and whatever alternate vehicles can be hastily repurposed. The scramble would be serious.

Scenario C — Mature adoption (50% car-free households, ~1,500 robotaxis). This is a city that has genuinely embraced shared mobility. Private vehicles handle approximately half the population. The robotaxi fleet can evacuate 6,000 people — 6%. Roughly 44% of the population — close to half the town, or 44,000 people — has no guaranteed evacuation vehicle. Emergency evacuation doctrine has no credible answer for this.

Scenario D — Full adoption (100% car-free, ~4,000 robotaxis). While unlikely to ever reach this degree of diffusion, it is worth contemplating as an end-point. There are no private vehicles. The entire commercially-sized fleet, loaded at emergency occupancy, evacuates approximately 16,000 people — 16% of the population. The remaining 84,000 residents are stranded, entirely dependent on emergency services that are also operating in the same degraded, panic-driven conditions.

While Scenario D may never happen, the others must be considered. Those are not worst-case numbers. They use generous assumptions about fleet size, and emergency vehicle occupancy.

The ownership baseline matters

U.S. Census data gives us a useful starting point. In genuinely rural and small-town communities — the ones most historically exposed to wildfire and flood evacuation — around 95% of households own at least one vehicle. In mid-sized cities with some transit infrastructure, that figure falls to roughly 90%. In the densest urban cores with robust transit — New York, Boston, San Francisco — it falls further still, though even in the New York metropolitan area, approximately 70% of households retain a vehicle.

The scenarios above are hypothetical, but they are grounded in this real gradient. Robotaxi adoption will not be uniform, and neither will evacuation risk.

The structural problem

The core difficulty is that the economics of shared fleet operation are structurally misaligned with emergency preparedness. A privately owned vehicle sitting in a driveway is, from a normal efficiency standpoint, a waste — it is idle roughly 95% of the time. But that idle asset is precisely what makes it useful in an emergency. It is available, instantly, without negotiation, coordination, network connectivity, or any operator making a dispatch decision.

When a city transitions toward Transportation-as-a-Service, it trades distributed, redundant, privately-held evacuation capacity for a centralized, commercially-optimized fleet that has no obligation or economic motivation to size for catastrophe.

What policy could address this

The transition away from private vehicle ownership will not happen overnight, and the scenarios above are illustrative rather than predictive. But they suggest that somewhere between 15% and 30% car-free households marks a threshold beyond which emergency evacuation risk begins to outpace the capacity of any commercially-sized robotaxi fleet coupled with an affordable emergency plan to compensate — even under optimistic loading assumptions.

Critical interventions deserve serious exploration:

• Emergency requisition frameworks could pre-designate the entire robotaxi fleet, along with school buses, transit vehicles, and commercial delivery fleets, as civil emergency assets — with pre-negotiated compensation and pre-programmed dispatch protocols.

• A municipal cap on the size of robotaxi fleets in communities with elevated natural hazard exposure — wildland-urban interfaces, floodplains, dam breach zones — effectively ensuring that such communities maintain household fleets, even as denser urban cores move aggressively toward car-free models.

• The inclusion of neighbouring robotaxi fleets into emergency response networks. This would move neighbouring fleets in advance toward emergency areas. This would be complex, require continuous planning, and not be workable everywhere.

• Mandatory reserve requirements, analogous to those imposed on electrical utilities, could require fleet operators above a certain market share to maintain vehicles beyond their commercial optimum. Even if acceptable, this would likely help very little.

• Municipal emergency vehicle stockpiles — a modest reserve held specifically for disaster deployment, not daily use. This would likely be unaffordable.

The issue worth addressing now

A privately owned vehicle sitting in a driveway looks like waste — idle most of the time. But that distributed, privately-held asset is precisely what makes mass self-evacuation possible. Replacing it with a centralized, commercially-optimized fleet means gaining efficiency on almost all days while surrendering resilience on the day that matter most. That trade-off deserves explicit policy debate, not just the implicit bet that catastrophic evacuations will be unaffected by declining vehicle ownership.

The time to think about this is before the transition, not during it.

—

Bern Grush is co-founder and CEO of Pudocity Inc. and Executive Director of the Urban Robotics Foundation. He co-authored The End of Driving (Elsevier) and contributes to ISO standards on vehicle-to-infrastructure and curb orchestration.

[1] Boswell, M.R. & Riggs, W. (2017). Could Autonomous Vehicles Save Lives in Disasters? Planetizen, November 15, 2017.