Waymo's Compute Lead Lays Out Why Camera-Only Self-Driving Stalls Past 99.9%
At Hot Chips, Waymo compute lead Daniel Rosenband gave a keynote on the challenges of autonomous driving compute.
1 source 1d ago
Reporting from 1 source: ASCII.jp.
Why it matters
Waymo's own framing is that more miles and more cameras do not buy the last decimal places, which is the argument behind building a Sensor Fusion Processor rather than following Tesla's camera-only route.
Key Facts
- Waymo compute lead Daniel Rosenband gave a Hot Chips keynote titled "Compute in Motion: Challenges of Autonomous Driving."
- Rosenband argued that camera-only autonomous driving can reach a certain level of results early but grows little beyond that point.
- Rosenband said driving ten times as many kilometers does not move a system from 99% to 99.9%.
- Waymo reports more than 200 million miles driven, more than 20 million paid rides, and 15 cities covered up to 2020.
- Waymo is developing a Sensor Fusion Processor.
He argued that camera-only systems can reach a certain level of results early but struggle to move from 99.9% to 99.99%, and that driving ten times as many kilometers does not add the extra nines. Waymo is developing its own Sensor Fusion Processor.
Rosenband's keynote was not a walkthrough of the Sensor Fusion Processor itself. It was the case for why Waymo started building one. His example was a red balloon floating in front of a traffic signal, mistaken for a red light, and he noted that even Gemini still seems unable to judge correctly whether it is a signal or a balloon.
Waymo's figures to 2020: more than 200 million miles driven, more than 20 million paid rides, 15 cities. Rosenband's message is that physical AI is only the first 1% and the product is the rest.
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google