- That seems low. I would like to see the quantity of human interventions too.
- Low? Put it this way. You work in a company or go to class with 45 people. Do you really think that 14 crashes in the last ~9 months is low?
In terms of miles driven, which is the fair comparison, from the article:
> Electrek analysis found that the vehicles have traveled roughly 800,000 paid miles in that time period, amounting to a crash every 57,000 miles. According to the NHTSA, US drivers crash once every 500,000 miles on average.
- >“The new crashes include a collision with a fixed object at 17 mph while the vehicle was driving straight, a crash with a bus while the Tesla was stationary, a collision with a heavy truck at 4 mph, and two separate incidents where the Tesla backed into objects, one into a pole or tree at 1 mph and another into a fixed object at 2 mph.”
>Electrek analysis found that the vehicles have traveled roughly 800,000 paid miles in that time period, amounting to a crash every 57,000 miles. According to the NHTSA, US drivers crash once every 500,000 miles on average.
Jesus, 10x worse than the average US driver is crazy. There's serious variability in human driving demographics, with some subsets already being that much worse than other subsets, things like "What city do you live in" can bring that stat down significantly.
So, to all the people who insist that autopilot is "better than human", do you still believe it? Assume it drives as many unpaid miles, that's still 5x worse than average drivers, which includes things like alcoholics, people who consistently drive high, people who don't really know how to drive, and that person you see on the road eating a fucking meal and doing their makeup and reading a book, and drivers who are currently asleep.
I uh did not actually expect the data to be this bad. Statistically, we need more data, but this is concerning. Not slowly tapping into objects that cannot move is probably a good way to improve those stats. Did Tesla abandon the industry standard ultrasonic sensors everyone else studs their bumpers with? Many other cars in the industry can parallel park themselves through these sensors, and reliably do not hit parked objects.
- this is misleading because they're not comparing apples to apples.
the insurance companies are looking into all the details I'm sure to be able price the risk accordingly. Lemonade is putting their money where their mouth is and it's pricing FSD miles at 1/2 the rate of manned driving. that's because FSD gets 1/2 the number accidents per mile.
- I think the discrepancy here is that almost all these crashes would not have resulted in an insurance claim, e.g. backing into a pole at 1 mph -- this is not enough damage to report for an average driver.
That said, really bad numbers for an autonomous system which is supposed to be way better than humans.
- > that's because FSD gets 1/2 the number accidents per mile.
I call bullshit and I bet Tesla is quietly paying Lemonade.
FSD is primarily used on highways, and the accident rate on highways is significantly lower per mile which results in FSD appearing to have a lower accident rate per mile.
Meanwhile Musk has a trillion dollars riding on them hitting 10 million FSD subscribers[1], so (past behavior being the best predictor of future behavior) he's obviously going to be committing whatever chicanery is required for him to get that money.
[1] https://finance.yahoo.com/news/why-elon-musk-1-trillion-1647...
- Is Tesla insuring their robotaxis thru lemonade?
- > insurance companies
If it's so good then why doesnt Tesla eat the liability?
- > Jesus, 10x worse than the average US driver is crazy.
And don't forget that per driver-mile, the US is already not great. Roughly three times as many deaths per mile as the UK, and 7.5 times as many accidents of any sort per mile.
36 points by speckx 3 hours ago | 10 comments