Since abandoning its self-driving car research and development division several years ago, Uber has faced constant questioning about its fate in the future era of self-driving cars. However, Uber seems to have found a new strategy to break through: transforming itself into an "arms dealer" for the self-driving industry. Uber's Chief Technology Officer, Praveen Neppalli Naga, recently stated this at a public event...First time revealingThe company's long-term ambition is to equip its millions of vehicles driven by human drivers with sensing kits, transforming them into a "sensor grid" that continuously collects data in the real world, and then providing this valuable driving data to self-driving car companies and AI model training.
From "AV Labs" to "Millions of Motion Sensing Devices"
This ambitious plan is actually an extension of Uber's "AV Labs" plan, which was announced in late January of this year.
Currently, AV Labs is still in its early stages, primarily relying on a small fleet of dedicated sensor vehicles operated by Uber (independent of its driver network) to collect data. However, Praveen Neppalli Naga admits that installing the devices in a large number of driver vehicles is their "ultimate goal," and the main challenge at present is to clarify the regulations in each state and confirm the legal boundaries of data collection and sharing by the sensor components.
It's conceivable that if this plan comes to fruition, the sheer scale of Uber's millions of drivers worldwide, if even a small fraction of them were willing to transform their vehicles into "mobile data collection platforms," would far surpass the fleet size that any single self-driving car company (such as Waymo) could deploy.
Solving the biggest pain point of self-driving cars: Corner case data
Praveen Neppalli Naga bluntly pointed out that the bottleneck in the development of autonomous driving today is no longer the underlying technology, but "data".
For example, companies like Waymo have to spend huge sums of money to deploy fleets of vehicles to collect data from different scenarios in order to train their models. However, the problem for these companies is that they do not have enough capital to deploy so many vehicles to collect this highly fragmented, scenario-specific information.
This is precisely Uber's advantage: through its ubiquitous driver network, Uber can easily obtain "corner case" data on various weather conditions, times of day, and rare road conditions.
Building an "AV Cloud" to upgrade from a ride-hailing platform to self-driving car infrastructure
Uber has already partnered with 25 self-driving car companies, including London-based Wayve. Praveen Neppalli Naga revealed that Uber is building a system called "AV Cloud": a database richly labeled with sensor data that partners can query and use to train models.
In addition, partners can test their trained models in "shadow mode" on real Uber trips, simulating how the self-driving cars react to real road conditions without actually driving them on the road.
