Uber Wants To Turn Its Millions Of Drivers Into A Sensor Grid For Self-driving Companies

Uber has a semipermanent ambition that goes good beyond shuttling passengers: nan institution yet wants to outfit its quality drivers’ cars pinch sensors to soak up real-world information for autonomous conveyance (AV) companies — and perchance different companies training AI models connected physical-world scenarios.

Praveen Neppalli Naga, Uber’s main exertion officer, revealed nan scheme successful an question and reply astatine TechCrunch’s StrictlyVC event successful San Francisco connected Thursday night, describing it arsenic a earthy hold of a nascent programme nan institution announced successful precocious January called AV Labs.

“That is nan guidance we want to spell eventually,” Naga said of equipping quality drivers’ vehicles. “But first we request to get nan knowing of nan sensor kits and really they each work. There are immoderate regulations — we person to make judge each authorities has [clarity on] what sensors mean, and what sharing it means.”

For now, AV Labs relies connected a small, dedicated fleet of sensor-equipped cars that Uber operates itself, abstracted from its driver network. But nan ambition is intelligibly overmuch larger. Uber has millions of drivers globally, and if moreover a fraction of those cars could beryllium transformed into rolling data-collection platforms, nan standard of what Uber could connection nan AV manufacture would dwarf what immoderate individual AV institution could combine connected its own.

The penetration driving nan program, Naga said, is that nan limiting facet for AV improvement is nary longer nan underlying technology. “The bottleneck is data,” he said. “[Companies for illustration Waymo] request to spell astir and cod nan data, cod different scenarios. You whitethorn beryllium capable to say: successful San Francisco, ‘At this schoolhouse intersection, I want immoderate information astatine this clip of time truthful I tin train my models.’ The problem for each these companies is entree to that data, because they don’t person nan superior to deploy nan cars and spell cod each this information.”

Becoming nan information furniture for nan full AV ecosystem is simply a beautiful smart play, peculiarly considering Uber years agone abandoned its ain ambitions to build self-driving cars (a move that co-founder Travis Kalanick has publically lamented arsenic a big mistake). Indeed, galore manufacture observers person wondered if, without its ain self-driving cars, Uber mightiness 1 time beryllium rendered irrelevant arsenic AVs progressively outpouring up astir nan globe.

The institution presently has partnerships pinch 25 AV companies — including Wayve, which operates successful London — and is building what Naga described arsenic an “AV cloud”: a room of branded sensor information that partner companies tin query and usage to train their models. Partners, which Uber plans to much aggressively invest successful directly, tin besides usage nan strategy to tally their trained models successful “shadow mode” against existent Uber trips, simulating really an AV would person performed without really putting 1 connected nan road.

“Our extremity is not to make money retired of this data,” Naga said. “We want to democratize it.”

Given nan evident commercialized worth of what Uber is building, that positioning whitethorn not past long. The institution has already made equity investments successful galore AV players, and its expertise to connection proprietary training information astatine standard could springiness it important leverage complete a assemblage that correct now depends connected Uber’s thrust marketplace to scope customers.