6 ms·
From the article: "These YouTube videos underscored how important it is for drivers to actively supervise Tesla's new software. Over the course of three hours,
by Animats 6y ago
From the article: "These YouTube videos underscored how important it is for drivers to actively supervise Tesla's new software. Over the course of three hours, the drivers took control more than a dozen times, including at least two cases when the car seemed to be on the verge of crashing into another vehicle."
In a way, it's good that the Tesla system sucks so badly. If they had a disconnect rate of one per month, drivers would trust the thing.
The other guys, disengagements per 1,000 self-driven miles in California:
Waymo, 0.076, or one per 13,000 miles.
Cruise, 0.082, or one per 12,000 miles.
Not clear how many of those would have resulted in a crash, as opposed to just stopping. Probably not many, the California autonomous DMV reports indicate. US humans have a crash rate of one per 508,000 miles driven.
Anyway, Tesla does not have "full self driving". They have slightly better Level 2. Not even Level 3. You'd think that by now they'd have hands-off freeway driving totally automated to a better than human level of safety. But no.
- banana_giraffe 6y agoMakes me wonder: I have two driving modes, not autopilot, but not as aware as I could be, and then tracking everything I can because something odd is going on and I'm trying to avoid a bad scenario. Perhaps I'm a bad driver, but I suspect this is normal, kind of in a low-attention mode most of the time, but if the car a lane over starts to change lanes too close, you go into high-alert mode. I wonder if there are studies on how often humans need to do this sort of "disengagement" for lack of a better term? Not sure how you'd measure it, perhaps with more eye movement than normal? Dunno.
- supernova87a 6y agoYou really shouldn't trust those disengagement statistics much at all. They are totally up to the individual companies what counts as "disengagement", how to define it, and how they report them.
- modeless 6y ago> Not clear how many of those would have resulted in a crash Waymo just published an extremely detailed paper explaining exactly how many of their disengagements would have resulted in a crash, among other things. It's very interesting reading and shows how far Tesla has to go. Summary: https://www.theverge.com/2020/10/30/21538999/waymo-self-driving-car-data-miles-crashes-phoenix-google https://www.theverge.com/2020/10/30/21538999/waymo-self-driv... The paper: https://storage.googleapis.com/sdc-prod/v1/safety-report/Waymo-Public-Road-Safety-Performance-Data.pdf https://storage.googleapis.com/sdc-prod/v1/safety-report/Way...
- jiofih 6y agoThis is reporting on a limited beta release. California’s official numbers to compare are Tesla, 0.076 per 1,000 miles, or 1 per 13,219 miles That was from 2018 though - AP has received hundreds of updates in the meantime but they stopped testing in California.
- sidibe 6y agoIf you believe those numbers they got really incredibly unlucky in every video I've seen that it happened to be while the camera was rolling
- leesec 6y agoPeople will really keep saying the Tesla system sucks up until they solve autonomy lol. They're the farthest ahead and it's not even close.
- remus 6y agoSure, but it'd be nice if in the meantime Tesla didn't call it 'full self driving', because it's nothing of the sort. It's a pretty good lane assist that requires constant human supervision and only works in a small subset of environments.
- mlindner 6y agoWaymo system is limited to a single city that is fully mapped into the Waymo system. Cruise is limited to only work on highways (and not all highways) and is only SAE Level 2. So you're comparing Apples to Giraffes right now, let alone oranges.