Uber plans to turn driver fleet into sensor grid for autonomous-vehicle companies

Uber plans to eventually equip the cars in its human-driver network with sensors that can collect real-world data for autonomous-vehicle companies and other firms training artificial-intelligence models on physical-world scenarios. The company's chief technology officer, Praveen Neppalli Naga, disclosed the long-term ambition during an interview at TechCrunch's StrictlyVC event on Thursday. He described it as a natural extension of AV Labs, a program Uber announced in January that currently relies on a small, dedicated fleet of sensor-equipped cars the company operates itself. Naga said Uber must first understand how the sensor kits work and ensure regulatory clarity in every state before expanding the effort to the broader driver network. The initiative is built on the premise that data, not technology, is the limiting factor for autonomous-vehicle development. Companies such as Waymo need large volumes of real-world scenarios to train their models, but lack the capital to deploy enough vehicles to collect it. Uber already has partnerships with 25 autonomous-vehicle companies and is building what Naga calls an "AV cloud," a library of labeled sensor data that partners can query and use to train their systems. Partners can also run trained models in shadow mode against actual Uber trips to simulate performance without putting a robotaxi on the road. Naga said Uber's goal is not to profit directly from the data. "We want to democratize it," he said. The positioning comes years after Uber abandoned its own effort to build self-driving cars, a retreat that co-founder Travis Kalanick has publicly called a mistake. By supplying proprietary training data at scale, Uber could gain significant leverage over an autonomous-vehicle sector that still depends on its ride marketplace to reach customers.