-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Early autopilot safety data. Tesla’s Autopilot lowers probability of having an accident by 50%
- Thread starter dhanson865
- Start date
-
- Tags
- Autopilot Tesla Inc.
I'm surprised it isn't higher, but I imagine we are following a steep curve up and won't plateau until their is 1 accident per 10 billion miles travelled...hypothetically.
Elon should be careful here.
In some cases when I use Autosteer, my safety level rises because I'm being more attentive babysitting it than I would be with it off. This isn't to the credit of Autopilot, it's due to my mistrust of it in some situations.
yep it's additive.
risk of human drivers = x > risk of autopilot = y > risk of autopilot + human driver = z
It'll be safer to have risk z of human + autopilot for several years to come.
Some day down the road we'll have autonomous driving which will be safer than any of the choices above but until that is very very common people that like to drive will choose autopilot + their own right to override or drive as the primary with safety features ready to intervene if needed.
Those that don't like to drive will welcome autonomous driving and won't want to participate in the duo with autopilot.
But not all the autopilot drivers that make up those stats are like you. So even for those less careful than you (which is likely most) autopilot is still safer.
I'm surprised he didn't say accidents are down even when autopilot is off too, because the forward collision protection would still reduce/minimize the most typical rear end accident.
Front radar is such an obvious safety improvement -- it will be required in some EU countries soon,and should be in the US too. Likely to save more lives and prevent more damage than anything since airbags and third brake lights.
If autopilot is used mostly on freeways, those are some of the safest miles already. I don't suppose they are comparing to non-autopilot freeway miles.
Not necessarily. It's potentially only short-lived -- i.e. at some point I may give up on putting up with its week points. On some trips, I already have.
Johan
Ex got M3 in the divorce, waiting for EU Model Y!
Didn't see this write-up by Fred Lambert posted here on the forum (apologies if this is a double post).
Very interesting.
Tesla’s Autopilot lowers probability of having an accident by 50% based on early data, says Musk
The obvious objection of course is the selection bias: Autopilot sees higher use in "easy" traffic conditions while it can't be used properly in demanding conditions (where most accidents happen). But Musk is intelligent enough to probably be talking about some actual effect hat they have inferred and not just the statistical phenomenon from selection bias.
Very interesting.
Tesla’s Autopilot lowers probability of having an accident by 50% based on early data, says Musk
The obvious objection of course is the selection bias: Autopilot sees higher use in "easy" traffic conditions while it can't be used properly in demanding conditions (where most accidents happen). But Musk is intelligent enough to probably be talking about some actual effect hat they have inferred and not just the statistical phenomenon from selection bias.
Are you sure? I thought it was based on miles driven together with airbag deployments. I got the impression that Tesla was tracking miles driven with AP on and associated airbag deployments vs. miles driven without AP and air bag deployments for that case. That said, freeway driving, where AP is mostly used, is much safer than other driving cases. Thus I question the conclusion.
Don't they say most accidents occur within a few miles of your home? I thought the stage of automation was limited to only highway driving, which would put you outside a few miles in most cases I'm sure.
I don't know how accurate it is to compare highway accidents with autopilot vs all accidents without.
I don't know how accurate it is to compare highway accidents with autopilot vs all accidents without.
That may well be the case, and Tesla will still continue to collect data so we will hopefully see what effect that has on the accident rate. But at the same time, the AP software will continue to improve.
This first report on how AP effects accident rates (an "accident" being defined as the airbags going off, which I think is a reasonable definition) is quite encouraging. I think the AP accident rate will continue to decline as AP improves.
Currently Tesla drivers are more likely to use AP on freeways and open highways and not on surface streets near their house, but the data Musk described is still a useful measure of AP effectiveness.
Johan
Ex got M3 in the divorce, waiting for EU Model Y!
A much more interesting methodology would be if non-AP cars fed the mothership with data on their driving patterns (speeds, freeway percentage, geo-location etc) as well as some hard measure on accident rate (such as airbag deployment) and then that a cohort of such cars could be matched against a cohort of AP-enabled cars with the same driving patterns to check for hazard ratio/odds ratio of an accident occuring. And that would naturally have to include accident any time, with or without AP enabled. This would be a way to try to discern whether AP in itself reduces accident proneness.
Crash <> accident.
AP might stop a crash, but less likely to stop an accident. Accidents are: a blowout, deer strike, somebody blowing a red light in front of you, or other unexpected outcomes of driving a car. Hitting something because you were not paying attention is not unexpected.
Driving while not looking ahead isn't anymore an accident than firing a gun blindfolded in a crowded room and shooting someone is accidental.
You were operating a car in public, but where you were aiming it at wasn't very important to you.
AP might stop a crash, but less likely to stop an accident. Accidents are: a blowout, deer strike, somebody blowing a red light in front of you, or other unexpected outcomes of driving a car. Hitting something because you were not paying attention is not unexpected.
Driving while not looking ahead isn't anymore an accident than firing a gun blindfolded in a crowded room and shooting someone is accidental.
You were operating a car in public, but where you were aiming it at wasn't very important to you.
Airbag deployment = airbag deployment
That is the event being measures that is imprecisly described as accident or crash.
That is the event being measures that is imprecisly described as accident or crash.
Similar threads
- Article
