Uber is pivoting its strategy in the autonomous vehicle space, announcing the launch of AV Labs, a new division focused on collecting and providing real-world driving data to other AV developers. The move signals a shift for the ride-hailing giant, which previously pursued in-house self-driving technology before selling that arm to Aurora in 2020-a decision prompted by a 2018 incident involving a pedestrian fatality . Now,Uber aims to become a key infrastructure provider,offering crucial datasets to accelerate the advancement of robotaxis and advanced driver-assistance systems for companies like Waymo,Waabi and Lucid Motors.
Uber previously halted its internal self-driving car development following a fatal incident in 2018 and sold the division to Aurora in 2020. Now, through AV Labs, the company will deploy a fleet of sensor-equipped vehicles across various cities to gather driving data. This data will be shared with partners including Waymo, Waabi, and Lucid Motors, with specific contract details still being finalized.
The demand for comprehensive driving data is surging as the autonomous vehicle industry transitions from rule-based systems to reinforcement learning. Real-world data is essential for training AV systems to navigate complex and unpredictable “edge cases.” This strategic shift marks a significant change for Uber within the autonomous vehicle ecosystem, evolving from a developer to a key infrastructure provider.
“This is a sign that, like many leading AI labs, they’re realizing that ‘solving’ for edge cases is a game of volume,” explained Praveen Neppalli Naga, Uber’s Chief Technology Officer, in an interview with TechCrunch. He added that greater access to driving data sets can help robotaxi companies address challenging scenarios, citing a recent incident where a Waymo vehicle was observed improperly passing a stopped school bus.
Democratizing Data to Advance the Industry
Uber currently has no plans to charge for the data collected by AV Labs. “Our primary goal is to democratize this data, right? The value of this data and the advancement of our partners’ AV technology is far greater than any revenue we could generate from it,” Naga stated. The availability of robust datasets is a critical factor in accelerating the development and deployment of autonomous driving technology.
Danny Guo, VP of Engineering at Uber, emphasized the need to establish a foundational data base before defining a specific product model. “Because if we don’t do this, we don’t believe anyone else can,” Guo said. “So, as someone who has the potential to unlock the entire industry and accelerate the entire ecosystem, we believe we need to take on this responsibility now.”
Starting Small with Big Ambitions
AV Labs is beginning operations on a relatively small scale, currently utilizing a single Hyundai Ioniq 5 vehicle, though Uber isn’t limiting itself to one model. The team is currently hand-installing sensors, including lidar, radar, and cameras. “We don’t know if the sensor kit will hold up, but that’s the grit we have,” Guo joked. “I think it will take some time before we can say we’ve deployed 100 cars on the road to start collecting data. But the prototype exists.”
Partners won’t receive raw data. Once the AV Labs fleet is fully operational, Naga explained that the division will “process and refine the data to help tailor it to our partners.” This layer of “semantic understanding” will be leveraged by the driving software of companies like Waymo to improve real-time route planning for robotaxis.
Guo also mentioned the possibility of an intermediate step where partners’ driving software will be integrated into AV Labs vehicles to run in “shadow mode.” Whenever an Uber AV Labs driver makes a decision different from the AV software in shadow mode, Uber will flag it for the partner company. This process will not only help identify flaws in the driving software but also train models to drive more like humans and less like robots.
A Tesla-Like Approach with a Focused Strategy
This approach mirrors the strategy Tesla has employed over the past decade to train its autonomous vehicle software. However, Uber’s approach doesn’t currently match Tesla’s scale, which benefits from millions of customer vehicles on the road daily. Despite this difference, Uber remains confident in its strategy.
Guo stated that Uber aims to focus data collection based on the specific needs of autonomous vehicle companies. “We have 600 cities we can choose from. If a partner tells us a specific city they’re interested in, we can deploy [our cars] there,” he explained. Naga added that the company hopes to grow the new division to several hundred people within a year and intends to move quickly.
While Naga envisions a future where Uber’s entire ride-hailing fleet could be utilized for data collection, he recognizes the need to start somewhere. “From our conversations with partners, they’ve just said, ‘give us anything that will help.’ Because the amount of data Uber can collect far exceeds anything they can do with their own data collection efforts,” Guo concluded.
