News for hackers. Stuff that matters.
Re:Alarmed by Tesla’s self-driving test, state legislators d... by amelius on Sunday December 12, 2021 @12:28PM (#29531511)
The problem is that self-driving is almost entirely about edge-cases. And machine learning can optimistically only reach about 99% accuracy, so that last 1% of edge cases is where the real problem lies.
Re:Alarmed by Tesla’s self-driving test, state legislators d... by ChuckNorris89 on Sunday December 12, 2021 @12:30PM (#29531533)
Yeah, but the issues the Tesla seems to have in the video above (pedestrians crossing, rail tracks, bus lanes, parked UPS trucks) are hardly 1% edge cases, but common occurrences that are in every city.

And it wasn't even rush hour through some crowded European city like Rome with people on scooters zooming in every directions, in poor weather conditions at night, it was broad daylight in a city with barely any traffic.

If it can't handle basic city driving properly, then I'm sorry.

Re:Alarmed by Tesla’s self-driving test, state legislators d... by ec109685 on Sunday December 12, 2021 @01:34PM (#29532202)
Agree with you, but that area in downtown San Jose is very weird to drive around given their light rail system.

Not defending Tesla for all of those mistakes, but as a human, it’s easy to end up going down the wrong lane.

Re:Alarmed by Tesla’s self-driving test, state legislators d... by TigeriusKirk on Sunday December 12, 2021 @01:53PM (#29532384)
I've personally seen cars in those San Jose train lanes enough times to wonder just how common it is.
Re:Alarmed by Tesla’s self-driving test, state legislators d... by bittercynic on Sunday December 12, 2021 @12:40PM (#29531639)
I think there's disagreement about what 1% means - far less than 1% of my time spent operating a car involves interaction with a UPS truck, but more than 1% of my trips do.

Asking as someone with no first hand experience with these systems - do they complete 99% of trips with no disengagements or unexpected behavior, 99% of time driving, something else?

Re:Alarmed by Tesla’s self-driving test, state legislators d... by m4nu3l on Sunday December 12, 2021 @12:41PM (#29531651)
I'd be curious to know what definition of edge-case you are using. I think self-driving can't be entirely about edge-cases by definition (of edge-case).
Re:Alarmed by Tesla’s self-driving test, state legislators d... by brewdad on Sunday December 12, 2021 @12:50PM (#29531757)
Sure 99% of driving is pointing the car straight, keeping it between the lines and avoiding the car in front of you. That's easy and has been essentially solved by dozens of car makers.

The "edge" cases are pretty much everything else. There are a near infinite number of possibilities here and it is clear that no one is anywhere near close enough to covering them all. Stopping for a pedestrian when turning right is not some completely unforeseeable event.

Re:Alarmed by Tesla’s self-driving test, state legislators d... by m4nu3l on Sunday December 12, 2021 @01:02PM (#29531870)
I get it. However what matters the most in my view is the long term ratio of serious accidents in self-driving vs human driving. There is no need to cover all possibilities for that, right? The only problem with Tesla is IMHO that the name FSD is misleading.
Re:Alarmed by Tesla’s self-driving test, state legislators d... by amelius on Sunday December 12, 2021 @01:34PM (#29532210)
> long term ratio of serious accidents in self-driving vs human driving

Human driver as in: average driver, or best driver?

Re:Alarmed by Tesla’s self-driving test, state legislators d... by clairity on Sunday December 12, 2021 @12:49PM (#29531749)
you’re talking about the definition of a single edge case. you can argue about the exact proportion, but it’s entirely within the realm of possibility that real world conditions are dominated by an amalgamation of many different edge cases that together add up to many percentage points.