Uber’s AV Labs program aims to outfit human drivers’ cars with sensors to sell real-world training data to AV companies and AI model builders at massive scale.
Key Takeaways
CTO Praveen Neppalli Naga says the AV bottleneck is now data access, not technology; Uber sees this as its moat.
Currently AV Labs runs a small dedicated sensor-equipped fleet; equipping millions of Uber drivers is the stated long-term goal.
Uber’s “AV cloud” offers labeled sensor data and shadow-mode model testing against live Uber trips to 25 partner AV companies including Wayve.
Uber holds equity in multiple AV players, giving it leverage as both marketplace gatekeeper and proprietary data supplier.
Regulatory uncertainty around sensor kits varies by state, which Naga cited as the near-term blocker before driver rollout.
Hacker News Comment Review
Commenters broadly questioned whether generic Uber-collected sensor data is actually valuable to AV companies that already have years of their own proprietary data, noting Uber’s timing looks late.
The Tesla counterexample surfaced repeatedly: billions of miles of fleet data have not produced full autonomy, undercutting the “data is the bottleneck” framing from Naga.
A thread on driver awareness raised labor concerns: drivers are contributing data used to automate their jobs, with little apparent knowledge of collective options.
Notable Comments
@AndrewKemendo: An LA Uber driver told him he didn’t care his driving data trained his replacement and had never heard of collective bargaining.
@JumpCrisscross: Raises the core technical question of how much more valuable proprietary AV data is versus generic sensor input Uber could collect.
