The development of autonomous vehicles is increasingly becoming a problem of data rather than simply sensors or computing power. Autonomous systems need to learn from an enormous variety of situations, including the rare and unpredictable events that are difficult to reproduce in a test environment. Uber is responding to this challenge by launching AV Labs, a new team focused on turning the company's extensive real-world mobility operations into data that can support autonomous driving development. 

The idea is based on the scale and diversity of Uber's existing network. Millions of trips take place every hour across cities, suburbs, airports and other environments, creating a continuous stream of real-world journeys, pick-ups, drop-offs and unexpected situations. Uber sees this operational footprint as a potential source of the long-tail scenarios that autonomous driving systems need in order to become more robust.

This is particularly interesting because the long tail is one of the difficult problems in autonomous vehicle development. Most driving situations are relatively ordinary, but safety-critical systems also need to cope with unusual combinations of road layouts, traffic behaviour and unexpected events. Simulation can reproduce many of these scenarios, but high-quality real-world data can provide a different level of realism and can reveal situations that developers did not anticipate.

AV Labs will bring together specialists in data, machine learning, computer vision, systems and infrastructure. Its work is expected to cover data mining, simulation, validation and improvements across perception, prediction and planning. In other words, the goal is not simply to collect more video or sensor recordings, but to create a process in which operational data can feed back into the development and validation of autonomous systems. 

The initiative also reflects a broader change in how autonomous mobility is being developed. As machine-learning-based systems become more sophisticated, the ability to continuously collect useful data, identify relevant scenarios and test improvements could become a competitive advantage. Mobility platforms that already operate at large scale may therefore have an interesting role in the autonomous ecosystem even when they are not developing the vehicles themselves.

For autonomous driving, the road ahead may ultimately depend as much on the quality of the learning loop as on the vehicle's sensors. Turning everyday mobility into a source of structured knowledge could help shorten the distance between what autonomous systems encounter in the real world and what they are able to handle safely the next time.

UBER. Uber Is Building a Real-World Data Engine for Autonomous Driving. Uber [online]. 2 March 2026 [cited 2026-09-09]. Available from: https://www.uber.com/us/en/newsroom/accelerating-the-worlds-autonomous-future/
 
Picture: chat gpt