Uber looks to cash in on self-driving cars — but not by driving them

Uber wants to capitalize on the emergence of self-driving cars — not by handing the wheel to AI-powered drivers, but rather by tapping the mountain of potentially valuable data the rideshare company could collect in the billions of trips it handles every year.

Uber this week announced a new initiative to collect and analyze data from vehicle cameras and sensors for its robotaxi partners. The goal: to generate real-world driving data valuable to autonomous vehicle (AV) companies.

Uber told CBS News it will start the effort by working with its 50,000 global fleet partners — third-party individuals or companies that own multiple vehicles and manage drivers that register their vehicles with Uber. Fleet partners will begin outfitting these vehicles with customized sensor kits that track weather and road obstructions, according to an Uber spokesperson. 

Uber said the sensor kits will be exterior-facing, not inside the car, and will focus on the public road environment.

"We have this platform strategy, and this is about helping our partners and accelerating equitable access to safe [autonomous vehicles] around the world," the spokesperson said. 

Uber declined to disclose which of its more than 20 partners, including Waymo, are involved in the effort. Canadian robotaxi company Waabi on Wednesday announced it is partnering with Uber to deploy 25,000 robotaxis on the platform in a deal valued at $1 billion.

Uber previously collected real-world data with its autonomous vehicle partner, Nvidia, and already has vehicles on the road today that are collecting data through cameras, the rideshare company has said said previously. 

In 2020, Uber stopped developing its own autonomous vehicles and sold the company's unit to self-driving car startup Aurora. That deal followed the death of a pedestrian hit by an autonomous Uber in 2018.

Real-world training

Autonomous driving companies and researchers have largely relied on simulations and algorithms to predict real-world traffic and driving problems to develop their products. For example, researchers from the University of Michigan developed AI to simulate terrible drivers, reducing the costs and complexity of testing the technology.

Uber told CBS News that one of its goals is to track unpredictable events, like trash cans blowing into a roadway or a pedestrian suddenly appearing in the dark, that synthetic models are worse at predicting. 

"The biggest bottleneck to autonomy is no longer software or hardware — it's access to superior, real-world training data and models," Uber Chief Technology Officer Praveen Neppalli Naga told CBS News in a statement. 

Such "long-tail data," as Uber calls it, is potentially lucrative for self-driving players, given that the sector's commercial potential depends on consumers feeling safe in an AV.  It could also provide a new revenue stream for Uber, which eventually plans to charge its partners a fee for the rideshare company's data. 

"This is really something that we can offer to supercharge the advent of this technology… we're very bullish and excited about it because the data can be very valuable right now," the Uber spokesperson said. "AVs at scale are a huge, trillion-dollar opportunity for Uber."

Rough road ahead?

Zachary Greenberger, formerly the chief business officer at Uber rival Lyft, also sees opportunity in the convergence of AI and traffic. He is now the CEO of Nexar, which develops tools for capturing and analyzing autonomous driving data. But getting up to speed quickly is likely to prove challenging for Uber, Greenberger told CBS News. 

Greenberger also pointed out that fleet drivers — that is, Uber's initial target for the new technology — are professionals and less likely than an inexperienced driver to get into the "crazy situations" that produce data that simulations cannot, like a child unexpectedly rolling a ball into the street.

"[T]he reality is that the math is pretty brutal. They would need to deploy hundreds of thousands of sensors onto vehicles, and they would need to do it very quickly to be able to provide data to these companies in a way that would be useful."