Uber and Lyft are betting on self-driving cars to become profitable. But that may not happen, new research from MIT suggests. (UBER, LYFT)

Uber and Lyft lose a lot of money.

Combined, the ride-hailing firms lost some $3.8 billion in 2019 alone, according to the companies’ IPO filings.

Most investors, and many Wall Street analysts, are pinning the premise of ever turning a profit on self-driving cars. After all, paying drivers is a massive expense that could be all but eliminated with autonomous vehicles.

But new research from MIT has thrown cold water on the idea that robotaxis will save the ride-hailing operators much money, at least not without massive improvements to their algorithmic dispatching efficiencies.

“Although the cost proposition of autonomous taxis (ATs) may be improved by more closely matching supply with demand, we demonstrate that achieving maximum utilization would still leave ATs fiscally uncompetitive with conventionally driven vehicles (CDVs),” authors Ashley Nunes and Kristen Hernandez write in their paper.

Specifically, their findings — based on data from San Francisco, a single market — point to a cost between $1.58 and $6.01 per mile to operate autonomous vehicles with single occupants. That’s much higher than the widely used $0.40 (or less) per mile estimate, and higher than the average costs of personal car ownership (which is around $0.59 per mile, according to US’ Bureau of Transportation Statistics).

Read more: Lyft executive suggests drivers become mechanics after they’re replaced by self-driving robo-taxis

What’s more, Uber and Lyft currently pay drivers up to $2 per mile in some of the more expensive cities, like New York.

Here’s the researchers’ full breakdown of estimates AV operating costs:


It all comes down to utilization rates

Utilization rates are a key measure for ride-hailing operators. In short, the measure accounts for the fraction of time that a driver (or autonomous vehicle) is shuttling a fare-paying passenger.

Multi-passenger trips, like those of current Uber Pool and Lyft Line offerings, could help bring costs down and increase those utilization rates, but the increases in efficiency will need to be dramatic, the researchers found.

“In a single ridership model, we find capacity utilization rates would need to improve by nearly 100% and margins lowered by 37% for autonomous vehicles to achieve cost parity with their conventionally driven counterparts,” they write. “In a multiple ridership model, achieving cost parity requires a 30 percent increase in occupancy rates and a 75 percent increase were a stronger cost proposition offered to incentivize shared autonomous vehicle use over conventionally driven vehicles.”

Lyft did not immediately respond to a request for comment from Business Insider and Uber declined to comment.

For their study, the MIT researchers used a utilization rate of 52%, based on their San Francisco sample data. But even if those rates do reach 100% utilization, operating costs might still be more than a traditional vehicle.


“We find that while capacity utilization levies the greatest influence on fares, and current utilization rates leave room for improvement, achieving maximum utilization yields a cost proposition that is still higher than CDV,” they write.

That might be complicated even further given that most consumers expect a steep discount when sharing rides with other customers, according to UBS research.

“If passengers are willing to share utilization can improve even further,” the Wall Street bank’s autos group, lead by Colin Langan, said in a research note to clients this week. “Unfortunately, the potential for sharing is low as consumers tend to demand a large discount for the added trip length & discomfort.”

More Robo-taxi news:


self-driving cars

Chevron icon
It indicates an expandable section or menu, or sometimes previous / next navigation options.

Read More

Audi self-driving unit taps newcomer Aeva for its unique lidar

Audi’s self-driving unit has tapped a startup with a unique approach to lidar as it ramps up testing in Munich using a fleet of autonomous electric e-tron crossover vehicles.

Audi subsidiary Autonomous Intelligent Driving, or AID, said Wednesday it’s using lidar sensors developed by Aeva, a startup founded just two years ago by veterans of Apple and Nikon.

Aeva, a Mountain View, Calif.-based company started by Soroush Salehian and Mina Rezk, has developed what it describes as “4D lidar” that can measure distance as well as instant velocity without losing range, all while preventing interference from the sun or other sensors. Move past the 4D branding-speak, and the tech is compelling.

Lidar, or light detection and ranging radar, measures distance. It’s considered by many (with Tesla as one exception) in the emerging automated driving industry as a critical and necessary sensor. And for years, that industry has been dominated by Velodyne.

Today, there are dozens of lidar startups that have popped up with promises of technological breakthroughs that will offer lower-cost sensors with better resolution and accuracy than Velodyne. It’s a promise that is fraught with challenges, notably the ability to scale up manufacturing.

Traditional lidar sensors are able to determine distance by sending out high-power pulses of light outside the visible spectrum and then tracking how long it takes for each of those pulses to return. As they come back, the direction of, and distance to, whatever those pulses hit are recorded as a point and eventually forms a 3D map.

Aeva’s sensors emit a continuous low-power laser, which allows them to sense instant velocity of every point in the frame at ranges up to 300 meters, the company says. In other words, Aeva’s sensors can determine distance and direction, as well as speed of the objects coming to or moving away from them.

This is a handy perception feature for autonomous vehicles operating in an environment of objects that travel at different speeds, like pedestrians, bicycles and vehicles.

Aeva, backed by investors including Lux Capital and Canaan Partners, says its sensors are also unique because they’re “free” from interference from other sensors or sunlight.

It was this combination of long-range perception, instantaneous velocity measurements at cm/s precision and robustness to interferences that sold AID CTO Alexandre Haag on the Aeva sensors.

Aeva spent the past 18 months going through a validation process with Audi and parent company Volkswagen. This announcement confirms that Aeva has made it past a critical hurdle in Audi’s AV plans. Aeva’s sensors are already on Audi e-tron development vehicles in Munich. The automaker plans to bring autonomous driving to urban mobility services within the next few years.

Interference is possible and can cause a stream of random points on a 3D map if the lidar is pointed directly at the sun or if there are multiple sensors on the same vehicle. Lidar companies have instituted various techniques to prevent interference patterns; autonomous vehicle developers also account for potential interference problems from the sun and snow by creating algorithms to reject these kinds of outliers.

Still, Salehian argues that interference is a significant challenge.

When you talk about the challenge of building to scale and designing for mass scale, it’s not just about how easily they can be manufactured, Salehian contends. “It’s also about having these things work in unison together on a row. So when you’re talking about hundreds of thousands of these cars, that’s a big deal.”

Read More