Poor people stand to gain substantial health benefits from the arrival of autonomous vehicles. But a new study finds they’re least likely to afford the rides.

Affordability concerns surrounding self-driving technology mean low-income citizens will largely miss out on improved access to transportation and the chance to buy cars equipped with systems that prevent collisions.

The study, published this month by the American Journal of Public Health, deflates some longstanding optimism from industry executives and government leaders who have touted autonomous vehicles as a means for improving the health of U.S. residents.

Self-driving vehicles have been expected to add those benefits, at least in part, by providing more efficient links between poor people and their medical appointments. More than 3.6 million patients miss or delay non-emergency treatment each year in the U.S. because of a lack of transportation, according to the report.

“Even with universal health care, poor people are disproportionately less likely to access health care, because they can’t get there,” Ashley Nunes, one of the study’s authors, told Automotive News. “There’s been hope that this technology can be used to narrow the gap in health disparity. We find it can’t.”

Using San Francisco as a model, researchers compared the costs of a robotaxi ride with those of owning a conventional older vehicle in the city. Examining costs of vehicle financing, licensing, insurance, maintenance, fuel and more, they found that a self-driving taxi would cost $1.58 per mile in a best-case scenario. Costs associated with traditional ownership of an older vehicle were three times less, at 52 cents per mile.

Achieving cost competitiveness would require “the complete forfeiture of profit expectations” by commercial fleet operators, according to the report, titled “The Price Isn’t Right: Autonomous Vehicles, Public Health and Social Justice.”

Whether government officials might reduce that disparity by eliminating licensing costs or subsidizing AV rides remains an intriguing proposition.

Automated vehicles are viewed as a means to reduce the number of missed medical appointments. In Columbus, Ohio, for example, officials launched a one-year pilot project in February that uses 12-seat self-driving shuttles to ferry residents of the Linden neighborhood to medical appointments, a recreation center, transit center, grocery store and child-care options.

Columbus secured funding for the project as part of a $40 million grant from the U.S. Department of Transportation, which named the Ohio capital winner of its Smart City Challenge in 2016. Using shuttles to address high rates of infant deaths in Linden was a key component of the city’s proposal.

In practice, the project has been a reality check on the promise and potential of AVs. Technology limitations forced the city to redraw the planned routes for the two shuttles. Weeks after the pilot began in February, NHTSA ordered a temporary halt to the project after a rider slipped on the floor during a sudden stop.

Now comes further understanding of the broad costs associated with operating self-driving taxis.

“The real promise of AVs is safe, affordable mobility on demand,” said Nunes, a researcher who holds appointments at Harvard and MIT. “That’s the true promise. But is it safe? Safe for whom? Affordable for whom? That was the goal of this particular study. If we give poor people this shuttle, will it be OK? What’s equitable about pooling their rides? Nobody wants to pool a ride, let’s be upfront about that.”

He said transportation decisions poor people face are fraught with health implications. If they forgo robotaxi rides to health appointments in favor of conventional car ownership, that, too, carries a safety cost.

Poor people are less likely to afford cars equipped with safety systems that either mitigate or prevent collisions. Those crashes already drain $18 billion annually from public coffers, according to the study, roughly $156 for every household. Subsidizing AV transit for the poor may reduce those costs.

“State and federal budgets are already paying for crashes in one form or another,” Nunes said. “So there’s a public-spending case to be made here. It’s ‘do you prefer to pay for those crashes before or after?’ ”

Nunes co-wrote the study along with Kristen Hernandez, former research assistant at MIT and now a policy analyst at Securing America’s Future Energy, and Sam Harper, associate professor at McGill University in Canada.