Hacker News Reader

Comment by andreareina | Hacker News Reader

It sounds like the cars are intending to route through a Slow Street[1], and when it gets there it discovers it can't and so reroutes, and in these situations the only way is by going back the way they came, is that right? Waymo should be able to quickly annotate streets as no-entry so it's not clear to me what's holding up the fix.

[1] https://www.sfmta.com/projects/slow-streets-program

View on HN

re "holding up the fix", another possibility is that Waymo wasn't aware of the problem, right?

If the cars were able to---albeit slowly---cope with the closed streets through this kind of cumbersome queueing in a dead end and turning around, it seems possible that Waymo might not have been aware there was an issue.

Would a “number of reroutes from original” metric catch this or does that have too much noise?

Or even a “amount of time to complete trip” vs “predicted time”.

How many cars that try to go down a street and fail are required before waymo becomes aware of the issue? Isn't pattern matching supposed to be the providence of machine learning? There's a pattern here. Car tries to go down a street and is surprised by failure. Over and over and over again. Surprise should only be surprising the first few times.

A routing system that’s trying to get cars from point A to point B in the quickest amount of time should notice the delay and route around it. But if Google’s system is just sending the cars around in random paths to generate a lot of training data, not optimizing for lower drive time, maybe there’s no system in place to do that.

> How many cars that try to go down a street and fail are required before waymo becomes aware of the issue?

I’m this is a like a tech variation on “if a tree falls in the forest, and no one hears it, does it make a sound?”

Yes, exactly. Trees are falling in the forest and nothing is set up to hear it. Setting something up to hear it is trivial, this is digital and not analog. Get on it, Waymo.

> "We continually adjust to dynamic San Francisco road rules. In this case, cars travelling north of California on 15th Avenue have to take a u-turn due to the presence of 'slow streets' signage on Lake," the company said.

They're aware. When they were made aware it doesn't say, so maybe it's just recent. But the problem's been ongoing for a while it sounds like.

EDIT: The cars have a safety driver, i.e. there's a human in the loop. No excuse for Waymo to not know about this basically immediately.

Reply on news.ycombinator.com