At a Glance
- Uber’s new Uber’s new AV Labs division will collect and share autonomous vehicle data with over 20 partners.
- The initiative will deploy sensor-laden cars across 600 cities without charging partners yet.
- Data sharing aims to accelerate training for companies like Waymo, Waabi, and Lucid Motors.
Why it matters: The volume of real-world driving data Uber can gather is far larger than what most autonomous-vehicle firms can collect alone, potentially speeding breakthroughs in safety and performance.
Uber has announced a new division, Uber’s new AV Labs division, that will gather and share autonomous vehicle data with more than 20 partners. The company says it will deploy sensor-equipped cars across cities to provide the raw material needed to train self-driving systems. The move comes as the industry shifts from rule-based operation to reinforcement learning, where data volume is a critical factor.
The Data Gap in Autonomous Driving
Autonomous vehicle companies rely on real-world data to tackle edge cases that simulations miss. Even a decade of operation has left gaps, as shown by Waymo’s recent illegal passing of stopped school buses. Uber’s chief technology officer Praveen Neppalli Naga explained that a larger data pool could help partners pre-empt such problems.
> “Having access to a larger pool of driving data could help robotaxi companies solve some of those problems before or as they creep up,” Naga told News Of Philadelphia.
The physical size of a company’s fleet limits the amount of data it can collect. Uber’s plan is to leverage its ride-hail fleet to bridge that gap.
How AV Labs Will Operate
The division is starting small, with a single Hyundai Ioniq 5 that Uber says is not tied to a single model. Engineers are still installing lidars, radars, and cameras.
> “We don’t know if the sensor kit will fall off, but that’s the scrappiness we have,” Danny Guo, Uber’s VP of engineering, said.
Guo emphasized that the lab must first build a data foundation before finding a product market fit.
> “Because if we don’t do this, we really don’t believe anybody else can,” Guo said. “So as someone who can potentially unlock the whole industry and accelerate the whole ecosystem, we believe we have to take on this responsibility right now.”
The plan is to deploy more cars gradually, with a target of a few hundred vehicles within a year. The fleet will collect data in 600 cities, and partners can request data from specific locations.
Data Sharing and Partner Interaction
Partners will not receive raw data. Instead, the division will process and “massage” the data to fit each partner’s needs. This semantic layer will help companies like Waymo integrate the data into their path-planning algorithms.
> “Once the Uber AV Labs fleet is up and running, Naga said the division will have to massage and work on the data to help fit to the partners,” he explained.
An interstitial step will also be introduced: Uber will run a partner’s driving software in “shadow mode” on AV Labs cars. Any deviation between the Uber system and the partner’s system will be flagged.
> “This will not only help discover shortcomings in the driving software, but also help train the models to drive more like a human and less like a robot,” Guo said.
The approach mirrors Tesla’s decade-long data collection strategy, though Uber’s scale is smaller. Uber plans to target data collection based on partner needs rather than blanket coverage.
Comparison to Tesla’s Approach
Tesla has millions of customer cars on the road, providing a massive data stream. Uber’s AV Labs, by contrast, will start with a handful of vehicles and grow over time.
> “We have 600 cities that we can pick and choose from. If the partner tells us a particular city they’re interested in, we can just deploy our cars,” Guo said.
Despite the smaller scale, Uber believes the sheer volume of data it can collect will outweigh what partners can achieve on their own.
Future Plans and Scaling
Naga expects the division to grow to a few hundred people within a year. Uber also sees the possibility of eventually leveraging its entire ride-hail fleet for data collection.
> “From our conversations with our partners, they’re just saying: ‘give us anything that will be helpful.’ Because the amount of data Uber can collect just outweighs everything that they can possibly do with their own data collection,” Guo said.
The division will initially operate without charging partners, focusing on democratizing data rather than generating revenue.
> “Our goal, primarily, is to democratize this data, right? I mean, the value of this data and having partners’ AV tech advancing is far bigger than the money we can make from this,” Naga said.
Key Takeaways
- Uber’s Uber’s new AV Labs division will provide autonomous-vehicle companies with processed driving data from a fleet of sensor-equipped cars.
- The initiative targets 600 cities and plans to grow to a few hundred vehicles within a year.
- Partners will receive data that has been semantically processed and will be able to run their own software in shadow mode.
- Uber is not charging partners yet, emphasizing data democratization over revenue.
- The move reflects a broader industry shift toward reinforcement learning, where data volume is critical.
The launch of AV Labs marks a significant step in Uber’s strategy to support the autonomous-vehicle ecosystem while leveraging its existing ride-hail infrastructure.
