Ride-hailing giant Uber announces a three-way collaboration with chip giant Nvidia and Israeli artificial intelligence startup Autobrains to launch a second robotaxi program in Munich. With a different software approach, the technology bets on fast scalability and carmaker independence.    

Uber’s new deal combines Autobrains’ autonomous driving software, Nvidia’s Drive Hyperion Level 4 compute platform, and its own ride-hailing network. Three companies, three layers of the stack. And… none of them are car manufacturers.

Serious money

Uber does not build cars. It does not build self-driving software either. What it does is connect millions of riders with millions of drivers. Increasingly, it wants to cut the human driver out of that equation entirely.

The company’s pitch to the world of autonomous vehicle technology is simple: we do not need to invent the car. We just need to be the platform it runs on. That means striking deals with whoever is building the best software, and Munich is where the latest bet is being placed.

The tech partner in this case is Autobrains, an Israeli company that most people have never heard of but that has been attracting serious money. Backed by Toyota AI Ventures and Continental, one of the world’s largest car parts suppliers, it has raised more than $140 million and quietly filed more than 300 patents. It also signed a deal with Vietnamese carmaker VinFast. Now it has Uber.

From single to multiple

Most self-driving systems rely on a single, enormous AI model that tries to handle every possible road situation at once, from a child stepping off a curb to the nearest motorway exit. It is powerful, but it is expensive to build and even more expensive to run.

Autobrains takes a different approach: its system breaks down driving into specialist tasks, each handled by a separate software agent focused on a specific kind of decision. The idea is that a more modular system can run on standard car hardware, without the need for custom chips or exotic sensor setups. In theory, that makes it cheaper to scale.

Nvidia chip

The hardware beneath the system comes from Nvidia, whose Drive Hyperion platform is fast becoming the industry-standard compute layer for Level 4 autonomy (the point at which a car can genuinely drive itself, with no human backup required).

But what about the car Uber and Autobrains will use? With the proprietary chip approach, companies can use any car. The model should work with vehicles from any manufacturer, but a specific model hasn’t been disclosed yet.

Repeatable model

Munich is not a random choice either. It has dense city streets, fast suburban roads, a working tram network, and a regulatory environment that is cautious but not closed. It also has more automotive engineers per square kilometer than almost anywhere on earth.

The importance of the city is further illustrated by the fact that the Autobrains partnership is Uber’s second robotaxi. It also plans to begin Level 4 testing with Chinese autonomous driving firm Momenta in Munich this year.

What happens in Munich will not stay in Munich. The explicit goal is a repeatable model that scales to other cities and other markets. The point of the pilot is not to prove the car can drive itself. The point is to prove that the business model works. Munich has not yet given regulatory approval.