Uber has launched AV Labs, a new team focused on accelerating the autonomy ecosystem by unlocking the hardest problem in the field today – a data flywheel that covers real-world, long-tail driving scenarios.
As machine learning technology advances, autonomy is increasingly a data and modeling race. Progress no longer hinges on closed-course testing or simulated environments alone – it depends on learning from rare, messy, real-world scenarios that only appear at scale. These edge cases are difficult and expensive to capture, and they remain a major bottleneck to deploying safe, reliable autonomy, at scale, in a reliable and sustainable way.
Every hour, millions of real Uber trips take place across cities, suburbs, airports, restaurants and complex environments. These trips provide the operational coverage and real-world context needed to capture the high-quality, long-tail autonomous vehicle data that autonomy has been missing.
AV Labs aims to bring together experts across data, machine learning, computer vision, systems and infrastructure to turn real-world operations into high-quality data that helps autonomous systems learn faster and perform better.
The team is set to support Uber’s AV partners and accelerate progress across the ecosystem by building the core capabilities that power autonomy learning, from data mining and simulation to validation and system-level improvements across perception, prediction and planning. Together they’ll harness Uber’s scale and real-world data to accelerate the future of transportation.
Related news, WeRide Genesis simulation model combines physical and generative AI