Uber will launch an ‘AV Labs’ unit to collect data on driving robotaxis

Uber has more than 20 autonomous vehicles, and they all need one thing: data. That’s the company he says will make this available through a new division called Uber AV Labs.

Despite the name, Uber is no returning to manufacturing its robotaxis, which it stopped doing after one of its test vehicles killed a pedestrian in 2018. (Uber eventually sold the division in 2020 in a problems with Aurora.) But it will send its cars to cities equipped with sensors to collect data for partners such as Waymo, Waabi, Lucid Motors, and others – although no contracts have been signed yet.

To put it simply, self-driving cars are in the midst of a transition away from a mandated role and rely more on motivational training. When this happens, the global movement data has become very important for training these systems.

Uber told TechCrunch that the independent companies that want this the most are the ones that have already collected data. It’s a sign that, like many frontier AI labs, they’ve realized that “solving” the most dangerous cases is a game of volume.

Physical boundaries

Currently, the size of the private sector’s fleet creates a limit to the amount of data it can collect. And although many of these companies create simulations of real-world situations to fight side-by-side cases, nothing beats driving a car on real roads – and driving a lot – when it comes to realizing all the strange, difficult, and unpredictable situations that cars are capable of.

Waymo exemplifies this difference. The company has had autonomous vehicles in operation or testing for a decade, but its latest robotaxis has caught on. passing illegally stopped school buses.

Having access to driver information could help the robotaxi industry solve some of these problems before they happen or when they start, Uber chief technology officer Praveen Neppalli Naga told TechCrunch in an exclusive interview.

And Uber isn’t charging. At least not yet.

“Our goal, basically, is to democratize this, right? I mean, the value of this data and the advancement of AV technology is greater than the money we can generate,” he said.

Uber’s VP of engineering Danny Guo said the lab needs to create a baseline before identifying the product’s potential market. “Because if we don’t do this, we don’t believe anyone can,” Guo said. “So as someone who can open up the entire business and accelerate the entire ecosystem, we believe we have to take this responsibility right now.”

Devices and sensors

A new phase of AV Labs is getting underway. Currently, it has only one vehicle (Hyundai Ioniq 5, although Uber says it is not married to a single model), and Guo told TechCrunch that his team would be rocking sensors like lidars, radars, and cameras.

“We don’t know if the sensor equipment will fall off, but that’s the problem we have,” he said with a laugh. “I think it will take time to say, put 100 cars on the road to start collecting data. But the model is there.”

Affiliates do not receive unedited data. Once the Uber AV Labs fleet is up and running, Naga said the team “needs to massage and use that information to help their partners.” This “semantic” information is what software companies such as Waymo will draw on to plan real-time robotaxi plans.

Even then, Guo said there will be an integration process, where Uber will plug its partner’s driver software into AV Labs’ cars to drive “in the shadows.” Every time an Uber AV Labs driver does something different from what the autonomous car software does in the shadows, Uber suspends the company.

This will not only help find errors in the driving software, but also help train models to drive more like a human and less like a robot, Guo said.

The Tesla method

If this method sounds familiar, that’s because it’s exactly what Tesla has been doing in training its car program for the past decade. Uber’s transportation is not at the same level, because Tesla has millions of customer cars that drive on the roads of the world every day.

This doesn’t bother Uber. Guo said he hopes to do more data collection based on the needs of independent companies.

“We have 600 cities that we can choose (from). If a friend tells us the city he wants, we can just send our (cars),” he said.

Naga said the company hopes to expand the new segment to a few hundred people within a year, and that Uber wants to move faster. And while he envisions a future where Uber’s entire fleet of snowmobiles can be leveraged to get more training, he knows the new phase has to start somewhere.

“Talking to our loved ones, they are just saying: ‘Give us anything that will help us.’ Because the amount of data that Uber can collect is far beyond anything they can do to collect their data,” Guo said.