Uber wants to turn millions of drivers into a sensor grid for self-driving companies

Uber has long-term ambitions that go far beyond picking up and dropping off passengers. The company ultimately wants to absorb real-world data for self-driving vehicle (AV) companies by equipping cars with human drivers with sensors to help them and potentially other companies train AI models in physical-world scenarios.

Uber Chief Technology Officer Praveen Neppalli Naga revealed the plan in an interview Thursday night at TechCrunch’s StrictlyVC event in San Francisco, describing it as a natural extension of an earlier program the company announced in late January called AV Labs.

“That’s ultimately where we want to go,” Naga said of equipping vehicles with human drivers. “But first you need to understand sensor kits and how they work. There are some regulations. We need to make sure all states are compliant with their regulations.” [clarity on] What does a sensor mean and what does it mean to share it? ”

For now, AV Labs relies on a small, dedicated fleet of sensor-equipped vehicles that Uber operates in-house, separate from its driver network. But the ambition is clearly much greater. Uber has millions of drivers around the world, and if even some of those cars can be turned into a rolling data collection platform, the scale that Uber can provide to the AV industry will dwarf the scale that individual AV companies can assemble on their own.

According to Naga, the insight driving this program is that the limiting factor in AV development is no longer the underlying technology. “The bottleneck is data,” he said. “[Companies like Waymo] You have to go around and collect data and collect different scenarios. In San Francisco, you might be able to say, “At this school intersection, I need data at this time of day so I can train a model.” The problem for all these companies is access to that data. Because they don’t have the capital to deploy cars and collect all the information. ”

Becoming the data layer for the entire AV ecosystem is a very smart strategy, especially considering that Uber abandoned its own ambitions to develop self-driving cars years ago (a move co-founder Travis Kalanick has publicly lamented as a big mistake). In fact, many industry observers wonder if Uber will one day become worthless without its own self-driving cars as more and more self-driving cars emerge around the world.

The company currently partners with 25 AV companies, including London-based Wayve, to build what Naga calls an “AV cloud,” a library of labeled sensor data that partner companies can use to query and train models. Partners, which Uber plans to invest in more aggressively directly, can also use the system to run trained models in “shadow mode” against real Uber rides, simulating how AVs behave without actually putting them on the road.

“Our goal is not to make money from this data,” Naga said. “We want to democratize.”

Given the obvious commercial value of what Uber is building, that positioning may not last long. The company already has equity investments in a number of AV players, and its ability to provide proprietary training data at scale could give it significant leverage in a space that currently relies on Uber’s ride-hailing marketplace for customer acquisition.