Uber has quietly put sensor-covered vehicles back on public roads, seven years after one of its self-driving test cars killed a pedestrian in Arizona. The new programme, called AV Labs, is not a return to the robotaxi business. The cars carry no passengers. They exist to record real-world driving data and feed it to the growing list of robotaxi companies that Uber partners with.
The crash context is worth holding in mind. In March 2018, an Uber autonomous test vehicle struck and killed Elaine Herzberg as she crossed a street in Tempe, Arizona. It was the first recorded pedestrian death involving a self-driving car. Uber halted all testing, lost its permits in Arizona and California, and eventually sold its entire self-driving division to Aurora in 2020. The company that swore off self-driving is now back with sensor hardware on public streets. The difference is that this time, someone is always driving.
So what is AV Labs actually doing? The cars are fitted with cameras, radar, and lidar, which are the distance-sensing instruments you see mounted on the roofs of robotaxis. A human driver operates the car normally. The sensors record everything: unusual traffic situations, complex junctions, pedestrians behaving unpredictably, bad weather, anything a self-driving system might struggle with. That recorded data gets cleaned, labelled, and handed to Uber's robotaxi partners to train their software.
There is a second feature called shadow mode. A partner's driving software runs silently in the background while a human drives. Every time the human does something different from what the software would have done, the system flags it and sends that gap back to the partner. It is a way of stress-testing software without putting a fully autonomous vehicle on the road. Uber is essentially running a quality-control service for the robotaxi industry.
The business logic is straightforward. Every robotaxi company, no matter how well-funded, is limited by the size of its own fleet. More cars means more data, but building and operating a large fleet is expensive. Uber already operates across 600 cities, with millions of trips happening every day. No single AV company can collect data at that geographic spread. Uber's CTO has been direct about this: the company believes the volume of data it can gather simply exceeds what any partner could collect on its own.
Uber's partner list has grown fast. It now includes Waymo, Volkswagen, May Mobility, Pony AI, Baidu, and others, reaching 25 companies in total. In some cases, Uber is also taking direct financial stakes: it committed a $300 million equity investment in Lucid as part of a robotaxi programme, and put $100 million into WeRide. The data business and the investment strategy are two sides of the same positioning: Uber wants to be the essential layer underneath the robotaxi industry, regardless of which company's car ultimately picks you up.
The longer-term ambition is considerably larger than one Hyundai Ioniq 5. Uber has indicated it wants to eventually equip its millions of regular human drivers' cars with sensor kits. Every Uber ride anywhere in the world could become a data-collection event. The regulatory picture for that is still being worked out state by state, but the direction is clear.
For any business that moves goods or people, the relevant signal here is not AV Labs itself. It is what AV Labs reveals about the pace of the broader industry. The robotaxi companies are hungry for data precisely because they are pushing hard toward large-scale commercial deployment. Uber is building infrastructure to accelerate that. The timeline for robotaxis arriving in more cities, in more countries, just got a dedicated accelerant behind it.