[smallmancontrov]

Their decision to rely on vision got them billions of miles of training data from every corner of the USA. It did not get them point clouds, but monocular depth estimation works extremely well these days, so would it really have been that valuable? It looks to me like they made a shrewd bet and won big.

[seanmcdirmid]

They didn’t count on LIDAR falling in price so rapidly with their bet.

[smallmancontrov]

The cameras that collected those billions of miles of data were installed up to 10 years ago. You can't teleport today's LIDAR prices into the past.

[seanmcdirmid]

Yes, but it’s pretty easy to drive a car around with a LiDAR unit, you can couple it with Google maps data collection.

[smallmancontrov]

They didn't drive around a collection car, they had their customers drive millions of collection cars for billions of miles.