Uber announced it is launching AV Labs, a new team focused on speeding progress across the autonomous vehicle ecosystem by tackling what it calls one of the toughest constraints in the field: building a scalable data flywheel that captures real-world, long-tail driving scenarios.
In the post, Uber frames autonomy as increasingly driven by data and modeling, arguing that progress depends less on closed-course testing or simulation alone and more on learning from rare, messy events that surface in real-world operations. Those edge cases, Uber says, are difficult and expensive to capture and remain a bottleneck to deploying safe, reliable autonomous systems at scale.
Uber positions its core advantage as operational scale and environmental variety. With millions of trips taking place each hour across dense cities, suburbs, airports, restaurants, and other complex settings, the company says it can observe a breadth of real pickup and drop-off dynamics, routing decisions, and unpredictable conditions across times of day and weather that few others can match at comparable capital efficiency.
According to the announcement, AV Labs will be a lean, high-velocity, multidisciplinary group spanning data, machine learning, computer vision, systems, and infrastructure. The goal is to translate real-world operations into high-quality datasets that help autonomous systems learn faster and perform better.
Uber says the team will support its AV partners by building core capabilities that power autonomy learning, including data mining, simulation, validation, and system-level improvements across perception, prediction, and planning. The company also notes that AV Labs is hiring engineers and researchers to work close to what it describes as the learning core of autonomy.
