Uber wants to turn millions of drivers around the world into the company's "sensor network"

Years after ending its self-driving project, Uber is trying to get back into the driverless car scene in another way: transforming the vehicles of millions of online ride-hailing drivers around the world into mobile "sensor arrays" that provide data to self-driving companies and other real-world AI models.

Uber Chief Technology Officer Praveen Neppalli Naga disclosed the long-term vision in an interview, describing it as a "natural extension" of the company's new project AV Labs announced at the end of January this year. He said that Uber’s ultimate direction is to install various sensors on human drivers’ private cars in the near future to collect real road scene data. Naga also emphasized that before taking this step, the company needs to thoroughly understand the capabilities and working methods of different sensor kits, and wait for U.S. states to give clearer regulatory guidance on "what is a sensor and how to share data."

Currently, AV Labs still operates on a limited fleet of dedicated vehicles equipped with sensors that are operated by Uber itself and are independent of the group of drivers who take daily rides. But it can be seen from Uber’s narrative that this is just a starting point: Uber has millions of drivers around the world, and even if only a small number of vehicles are equipped with sensors, it is enough to build a road data collection network that is difficult for any single self-driving company to match. Naga believes that the bottleneck restricting the evolution of autonomous driving technology is no longer the underlying algorithm or computing power, but high-quality and sufficiently diverse real-world data. “The bottleneck is data,” he said. “Companies like Waymo need to constantly go out and collect data to cover different scenarios.”

In his vision, self-driving companies can customize extremely detailed training data on demand through Uber's network, such as the requirement to "collect traffic conditions at an intersection in front of a school in San Francisco during a specific time period to train the model." The real problem is that most autonomous driving companies do not have sufficient capital to deploy their own fleets on a large scale around the world to cover these long-tail scenarios with high density. If Uber can mobilize existing driver and vehicle resources, it is expected to become the data supply layer for the entire industry, providing a steady stream of "fuel" for autonomous driving technology.

The outside world has long questioned whether Uber will be "bypassed" by self-driving companies in the future after giving up building its own self-driving cars, or even be marginalized in the travel ecosystem. Co-founder Travis Kalanick has also publicly stated that giving up autonomous driving is a "huge mistake." Today, through AV Labs, Uber is trying to transform its role from a self-driving vehicle developer to an infrastructure and data platform in this field, providing underlying capabilities to all participants with the help of its extensive driver network and order flow.

Uber currently has partnerships with 25 self-driving companies around the world, including players such as Wayve, which operates in London. On this basis, the company is building a so-called "AV cloud": a fully annotated multi-modal sensor data warehouse that partners can retrieve and call to train their own autonomous driving models. Naga said that partner companies can also run "shadow mode" inference on real orders on the Uber platform - that is, simulating how their own autonomous driving systems will make decisions on real trip data without actually putting autonomous vehicles on the road.

Judging from its external appearance, Uber is trying to package this platform as an "industry public facility." "Our goal is not to make money from this data," Naga said, "but to democratize it." However, given the commercial value and scarcity of high-quality data in autonomous driving and the broader AI field, whether such a positioning can be sustained in the future is still questionable. In fact, Uber has made equity investments in a number of self-driving companies in recent years, and if the large-scale and differentiated training data it possesses becomes part of its partners' core competencies, Uber's bargaining power in front of these companies is likely to be further strengthened.

Behind this idea, Uber's logic is shifting from "building a car" to "building a platform": on the one hand, it continues to maintain its entry advantage at the end-user level through its own travel and food delivery network; on the other hand, it tries to precipitate the real itinerary and scenes of the driver's vehicle into structured data assets to serve autonomous driving companies and even other large model companies that require training data from the physical world. For a company that has long since stopped making autonomous driving hardware and software stacks, this may be a new way to continue to participate in the next round of transportation technology changes and maintain a presence in it.