Hacker News Reader

Comment by jefftk | Hacker News Reader

I think there's some chance that Tesla's approach will work out for them, but I'm not optimistic. Getting self-driving software to "impressively good" happened pretty quickly, and even systems from 2016 would usually pass your "go there without touching the wheel or pedals" test. But from there to "as safe as a human" let alone "as safe as we require human-replacing machines to be" turns out to be quite a hard gap to cross.

View on HN

There’s no comparison between what was available when I bought my Tesla in 2020 and what I have today.

When you say "what was available" do you mean what a Tesla was capable of? Because I was trying to talk about things like Cruise driving for 90min at night in a city with no intervention: https://youtu.be/KSRPmng1cmA

If you let me pick the roads, I could easily drive 90 minutes at night with no intervention in the LA area: driving from Ventura, down the 126 to the 5 to LAX can be done basically with no interventions no problem today. It could be done with EAP two years ago too. But, FSD works well on essentially arbitrary roads and it didn't four years ago.

> driving from Ventura, down the 126 to the 5 to LAX

That's a lot simpler than what Cruise showed in their video. Dealing well with pedestrians, stopped cars, cross streets, etc are a significant challenge.

That basically works today, my point is that there was a step change in ability over the last year or two compared to when I got my car.

> systems from 2016 would usually pass your "go there without touching the wheel or pedals" test

Can you give an example? I'm not aware of anything in 2016 that would consistently pass this test.

Well, there's consistently and there's consistently.

In 2016 Waymo reported their safety drivers intervening once for every 5,100 miles driven [1] - which implies to me that 99% of journeys nobody touched the wheel or pedals.

The problem is 99% isn't enough, as there are tremendous numbers of cars out there, and a busy bit of freeway would get a disengagement per minute.

[1] https://driverless.wonderhowto.com/news/2016-disengagement-r...

They Waymo trick is entirely different to FSD. They are travelling pre-mapped streets.

with remote controller drivers on an as needed basis

You've used Google Street View, right?

It might be missing a few logging roads, private driveways and carpark lanes - but it shows Google is entirely capable of mapping streets in detail all over the world.

> and even systems from 2016 would usually pass your "go there without touching the wheel or pedals" test

This is complete bullshit. The systems in 2016 were restricted to pre-mapped areas with lots of training. (i.e. the waymo/cruise approach today) So if you weren’t in a tiny slice of San Francisco or a few other training areas, this didn’t work.

Even if an area is pre-mapped, if you're operating on public streets you need to handle all sorts of unusual things. Here's a Cruise video recorded in 2016 on SF streets, showing a tricky interactions with buses, stopped delivery vehicles, pedestrians, etc: https://www.youtube.com/watch?v=1Tp6Ubf6mE4

Yeah, I’m not saying they weren’t doing impressive things. I’m saying they were still severely limited and would not be capable of a “go around without touching the wheel/brakes test” like a Tesla can in the entire US now.

Reply on news.ycombinator.com