Crash Rates

[EVNow]

Moving this discussion into a new thread.

"Nationally, 6.1 million miles of driving by a good driver should result in about 40-60 events, most of which are small dings, 22-27 or which would involve an insurance claim, 12 which would get reported to police and 6 injury crashes."

Waymo Data Shows Superhuman Safety Record. They Should Deploy Today

If we count "small accidents" as small dings and insurance claims that comes to 42 out of 6.1M miles. Doing the math, 6.1M divided by 42 gives us 1 per 145k miles.

So small accidents would be around 1 per 145k miles. So I think the 1 per 150k miles is about right.

Here is the data from Cruise. L1 being severe and L4 being small dings etc. So, about 10k miles per "crash" or "error" in FSD terms. Thats about 1 per year of driving - and seems about right. I've to hunt for the original source document - but these are the figures I've in my spreadsheets (along with lots of other data I've gathered).

Crash Type   Miles
L1           282,485
L2+           99,404
L3+           20,661
L4+            9,218

If we take 150k miles per even small dings - or more than 10 years of driving - that looks just wrong on the face of it. Just look around at all the cars - how many are a few years old and have dings ? Almost all of them. Ask yourself - how many times have you had small/big crashes in the last 10 years ?

 

[diplomat33]

Moving this discussion into a new thread.



Here is the data from Cruise. L1 being severe and L4 being small dings etc. So, about 10k miles per "crash" or "error" in FSD terms. Thats about 1 per year of driving - and seems about right. I've to hunt for the original source document - but these are the figures I've in my spreadsheets (along with lots of other data I've gathered).

Crash Type   Miles
L1           282,485
L2+           99,404
L3+           20,661
L4+            9,218

If we take 150k miles per even small dings - or more than 10 years of driving - that looks just wrong on the face of it. Just look around at all the cars - how many are a few years old and have dings ? Almost all of them. Ask yourself - how many times have you had small/big crashes in the last 10 years ?

I've driven ~20 years with only 1 real accident (losing control on icy road and going off the road, no other vehicles hit) and a couple of curb rashes. I've had some near misses though. But ultimately, that is anecdotal. Just because you've seen a lot of old cars with dings does not mean that the statistic is wrong. The stat is based on a larger sample.

I won't quibble over the stats. I am just quoting what the article says. The 1 accident per 150k miles stat is for more than dings though.

I would add that the article data and the Cruise data are not equivalent. Cruise data is looking at accidents by severity type. The article references accidents by "dings", "insurance claims", "police reports" and "injury crashes". So they are not using the same metrics at all.

I can accept your Cruise data. My main point still stands that 1 intervention per 378 miles, as the other poster claimed, is definitely not good enough. It will not get you better than human safety.

 

[EVNow]

I can accept your Cruise data. My main point still stands that 1 intervention per 378 miles, as the other poster claimed, is definitely not good enough. It will not get you better than human safety.

Ofcourse - 1 error per 378 miles won't fly. Per Cruise the current rates are 1 per 10k miles - so some 30 times better than that.

Now - FSD "error" is probably a little different from the crash rates. Essentially crash is something that happens because of an error - but not all errors result in crashes. We have all had a lot of near crashes - you change the lane and someone honks. I don't know what that rate is - I'll look around. Most of this kind of solid data comes from SHRP - The Strategic Highway Research Program run at Virginia Tech Transportation Institute. See the attachment for one such report.

Where AVs can really make a difference is by being attentive all the time. Something we can't do - and every time I've had any kind of crash (dings / tire scrapes and one rear-ending) - I've been inattentive / distracted. AVs can eliminate that.

 

[tsh2]

30 years driving (10-15k miles per year probably). Zero reported incidents. 2 minor moving contact (one from behind). 3 or 4 manoeuvring errors (half being selecting reverse rather than 1st gear, both hired vehicles). Maybe 10 near misses (didn't look/see, insufficient traction, etc), some of which had scope to be fatal. Maybe the last ones are interesting because they are mostly observational failures, and should count against FSD even if there is no trace of them in a human record. These are of course only the ones I noticed at the time and still remember well enough to count.

Based on this, one minor (non crash) per 10k miles for an average human seems reasonable.

 

[KITT77]

Daily I take the most dangerous freeway in America, the 99, and before that, Los Angeles' 5 freeway, I always had long 45 minute commutes to work. In my 28 years of driving, I had one major accident. I was sitting at an intersection and some lady plowed into me. Now I leave a good space in front of me at intersections and I leave the camera on -- thank God for the new update where, when the cameras go up on the screen on a model 3, the next turn in the directions move to the visualization screen on the left.

Point is, people evolve and because people evolve, while some don't, some drivers almost never get into accidents. Just always got to be aware. People are looking for FSD so much to take over, like it has to be something else driving. I see the driver intervention part as important, and even a time frame to learn to work with AI to be a better overall driver. I do see a point in having it. I do believe it expands my field of awareness, but I'm also not the kind that lets autopilot "do its thing." I never stopped making changes to my defensive driving, so beyond curb rash, and with a little luck, you're good.

 

[electronhauler]

Daily I take the most dangerous freeway in America, the 99, and before that, Los Angeles' 5 freeway, I always had long 45 minute commutes to work. In my 28 years of driving, I had one major accident. I was sitting at an intersection and some lady plowed into me. Now I leave a good space in front of me at intersections and I leave the camera on -- thank God for the new update where, when the cameras go up on the screen on a model 3, the next turn in the directions move to the visualization screen on the left.

Point is, people evolve and because people evolve, while some don't, some drivers almost never get into accidents. Just always got to be aware. People are looking for FSD so much to take over, like it has to be something else driving. I see the driver intervention part as important, and even a time frame to learn to work with AI to be a better overall driver. I do see a point in having it. I do believe it expands my field of awareness, but I'm also not the kind that lets autopilot "do its thing." I never stopped making changes to my defensive driving, so beyond curb rash, and with a little luck, you're good.

CA Hwy 99 is probably not the most dangerous road. Most serious collisions in the US occur on rural roads. Hwy 99 has it's share of trouble with fog and truck/vehicle collisions, nevertheless despite the high number of problems, it also has a very high 'vehicle miles travelled'. As with most freeways they are actually the safest roads to travel on in terms of vehicle miles per fatality or serious collision.
A third of fatalites have excess speed as a contributory factor. That equates to about 35 people per day in the US.