Falling asleep with Autopilot on -- does it explain Tesla's numbers?
Submitted by brad on Fri, 2019-09-13 12:21
If you fall asleep while driving with Autopilot, it's not a good idea, but you're much safer than falling asleep in a car that doesn't have it. Since a lot of accidents are caused by falling asleep, could this be the reason for Tesla's claim that driving with Autopilot is much safer than not? Or is that claim itself even valid? I outline the logic and math in my new Forbes site article at:
Sleeping with Tesla Autopilot may explain some of Tesla's numbers
Fri, 2019-09-13 15:45
This self reported stat by Tesla is complete bunk. Suggest looking at actual insurance numbers. Maybe check out the Tesla Death list, because in truth, Tesla has a much higher fatality rate than comparable luxury vehicles. You can view them at Tesladeaths.com.
Fri, 2019-09-13 16:11
Actual insurance numbers
Have these been released? I suspect Tesla is telling the truth about miles per accident, but they don't disclose what an accident is, so it's not possible to compare.
Sat, 2019-09-14 06:08
Lies and statistics
"[Tesla] drivers not using Autopilot should not have 3 times as good a record than average"
Probably not, but even that is hard to say without more data. I'm sure there is a significantly different rate of accidents between Tesla drivers and average drivers. Whether it's higher or lower, and how much, I can't say, though.
Probably not 3x. But that's comparing all Tesla drivers to average drivers. This statistic was comparing Tesla drivers "without Autopilot and without our active safety features." I'm not sure what group that would be, but they are likely to have much different driving habits than the average Tesla driver and the average driver.
Sat, 2019-09-14 10:32
The non-autopilot are mostly non-highway, while the NHTSA numbers are for all roads. Yes, Tesla drivers are older and richer.
But more to the point, while NTHSA has access to police records and can account for accidents based on police reports, I think that would be harder for Tesla, which is going to find it vastly easier to get reports from vehicles on impacts detected with accelerometers and airbag deployments or even crash repairs. They won't say what they are using.
Sat, 2019-09-14 16:57
Remember that not all cars have Autopilot enabled. For this reason (as well as the fact that not everyone uses Autopilot whenever it is available), I don't think it's safe to say that non-autopilot miles are mostly non-highway.
"without Autopilot and without our active safety features"
I took that to mean cars that were built before a certain point when Tesla began including "active safety features." Teslas built before September 2014 don't even have Autopilot as an option.
Tesla owners who bought before September 2014 are even richer, I assume. Whether or not that makes a 3x difference in accident rates, I can't say, though.
Sat, 2019-09-14 22:08
Apples to apples
It is hard to compare. I mean Tesla could do it, but they publish this statement to make it appear we should compare their numbers with the NHTSA ones.
The reality is that the NHTSA number includes all types of cars, all types of drivers, and all types of roads in all states. Even the Tesla non-ADAS number is going to have a focus on richer, older drivers and the states where electric cars are popular, in particular California.
Tesla could create a proper comparison. For their accidents, they know stuff like what road it was on, vehicle speed, and a bit about the owner if not the driver. They could pick a subset of their data that controls for the factors that are different.
However, comparing with the overall national average is not that useful. What they could do, very easily, is make the autopilot/non-autopilot comparison meaning by comparing the very same network of roads in the very same ratio. They could say, "In a mix of 50% California interstate and 50% urban streets of LA, SF, etc. here is how the two systems performed."
This might actually still tell the story Tesla wants to tell -- that Autopilot does better than non-autopilot. I have told them, why not do this? Why have it look like you are hiding something if you don't have anything to hide? It would help, not hurt, to provide real numbers which may not look as good as this but still look good.
Sat, 2019-09-14 20:22
I think the data in general is flawed anyway. I mean, as a whole, measuring car accident rates per mile driven is a relatively accurate way to score. However Tesla's are relatively new and there are alot less of them compared to all the others. If every Chevy drove one mile youd get a couple million miles right there, yet if every Tesla drove one mile you wouldn't even break a million.
Add new comment