3 ms·

Apparently Tesla has three billion miles of driving data. How many billion miles more are needed for the computer to not drive over a double yellow line? Is t

Apparently Tesla has three billion miles of driving data. How many billion miles more are needed for the computer to not drive over a double yellow line? Is there any evidence that three trillion miles would work? yall_got_any_more_of_that_driving_data_dave_chappel_meme.gif

socalgal2 1y ago
Waymo is working great with far less miles (40 million?). I use it all the time. And know many other friends who use it all the time and love it.
Zigurd 1y ago
3 billion miles of two dimensional 1080P images is what they've got plus telemetry for speed, acceleration, etc. A lot of the imagery was acquired at night or in foul weather or taken by a camera with a dirty or broken lens. If it turns out that lidar is necessary, there goes the value of the installed base of Tesla, vehicles. It would be like starting from scratch. Clinging to vision only AV is some variant of the sunk cost fallacy. If you're thinking, surely they must've thought of all that and have a plan, I'd point to the design of starship and ask, is that ever going to be human rated?
- BrandonLive 1y ago
  They collect video, not images, along with other sensor and control data. It’s not a sunk cost fallacy, it’s a technical strategy that is very logical and showing compelling results (though as of yet unproven for achieving robust L3 or L4 autonomy, demos notwithstanding).
BrandonLive 1y ago
Driving over a double yellow is expected and legal in normal driving, such as when making a left turn or going around an obstruction. In this example it looks like it oscillated between two different routing choices (turning left and going straight), and by the time it decided the correct route was to go straight, it found itself misaligned with the lane it should have been in. Instead of moving all the way back to the right, it kind of “cheats” its way into the upcoming left turn lane. This isn’t something it should do in this situation, but it’s likely emulating human behavior seen in situations like this which appeared in its training data, where people cut across the center line(s) ahead of the turn lane forming when they can see that it is clear. The thing a lot of people get wrong is that they think the most valuable data for Tesla to collect are the mistakes or interventions. Really what they need the most is a lot of examples of drivers doing good job of handling complex situations. One challenge, though, is separating out those good examples from the less good or bad ones, as human drivers are notoriously bad at, well, driving.

socalgal2 1y ago

Waymo is working great with far less miles (40 million?). I use it all the time. And know many other friends who use it all the time and love it.

Zigurd 1y ago

3 billion miles of two dimensional 1080P images is what they've got plus telemetry for speed, acceleration, etc. A lot of the imagery was acquired at night or in foul weather or taken by a camera with a dirty or broken lens. If it turns out that lidar is necessary, there goes the value of the installed base of Tesla, vehicles. It would be like starting from scratch. Clinging to vision only AV is some variant of the sunk cost fallacy. If you're thinking, surely they must've thought of all that and have a plan, I'd point to the design of starship and ask, is that ever going to be human rated?

BrandonLive 1y ago
They collect video, not images, along with other sensor and control data. It’s not a sunk cost fallacy, it’s a technical strategy that is very logical and showing compelling results (though as of yet unproven for achieving robust L3 or L4 autonomy, demos notwithstanding).

BrandonLive 1y ago

Driving over a double yellow is expected and legal in normal driving, such as when making a left turn or going around an obstruction. In this example it looks like it oscillated between two different routing choices (turning left and going straight), and by the time it decided the correct route was to go straight, it found itself misaligned with the lane it should have been in. Instead of moving all the way back to the right, it kind of “cheats” its way into the upcoming left turn lane. This isn’t something it should do in this situation, but it’s likely emulating human behavior seen in situations like this which appeared in its training data, where people cut across the center line(s) ahead of the turn lane forming when they can see that it is clear. The thing a lot of people get wrong is that they think the most valuable data for Tesla to collect are the mistakes or interventions. Really what they need the most is a lot of examples of drivers doing good job of handling complex situations. One challenge, though, is separating out those good examples from the less good or bad ones, as human drivers are notoriously bad at, well, driving.