Q3 2018 Vehicle Safety Report

[strangecosmos]

Tesla just released its Q3 2018 Vehicle Safety Report:

Here’s a look at the data we’re able to report for Q3:

• Over the past quarter, we’ve registered one accident or crash-like event for every 3.34 million miles driven in which drivers had Autopilot engaged.
• For those driving without Autopilot, we registered one accident or crash-like event for every 1.92 million miles driven. By comparison, the National Highway Traffic Safety Administration’s (NHTSA) most recent data shows that in the United States, there is an automobile crash every 492,000 miles. While NHTSA’s data includes accidents that have occurred, our records include accidents as well as near misses (what we are calling crash-like events).

Moving forward, we will publicly release these accident figures on a quarterly basis.

While this is a good start, we need more data. Tesla should compare the crash rate of miles driven with Autopilot engaged on freeways vs. the crash rate of miles driven with Autopilot not engaged on freeways.

It’s conceivable that the set of non-engaged miles would be biased towards more complex and more dangerous situations. So, another way would be to compare the freeway miles of customers who have purchased Autopilot and those who haven’t. Both metrics should be presented.

Also, HW1 Autopilot and Enhanced Autopilot statistics should be broken out independently.

 

[caligula666]

The report lacks depth to really understand what they mean. AP is less likely used on local and city roads, yet the likelihood of an accident per km is much greater than on highway where AP is typically used. So is this highway data only? This would imply cars would somehow internally have dedicated counters to track what distance was driven on what type of roads? Or that there is a huge bias.

 

[strangecosmos]

The report lacks depth to really understand what they mean. AP is less likely used on local and city roads, yet the likelihood of an accident per km is much greater than on highway where AP is typically used. So is this highway data only? This would imply cars would somehow internally have dedicated counters to track what distance was driven on what type of roads? Or that there is a huge bias.

Yes, I think there is a bias. Benchmarking Autopilot miles — predominantly driven on freeways — against all others miles is a skewed comparison. I believe most crashes occur off freeways.

The ideal would be for Tesla to turn over all its raw data to a trusted third party who can conduct an independent analysis.

 

[caligula666]

Agreed 100%

 

[strangecosmos]

I emailed Tesla and asked them if they would release more information. Worth a try.

 

[S4WRXTTCS]

Tesla just released its Q3 2018 Vehicle Safety Report:



While this is a good start, we need more data. Tesla should compare the crash rate of miles driven with Autopilot engaged on freeways vs. the crash rate of miles driven with Autopilot not engaged on freeways.

It’s conceivable that the set of non-engaged miles would be biased towards more complex and more dangerous situations. So, another way would be to compare the freeway miles of customers who have purchased Autopilot and those who haven’t. Both metrics should be presented.

Also, HW1 Autopilot and Enhanced Autopilot statistics should be broken out independently.

Agreed, but there also has to be a separation between TACC only versus AP.

My rationality is that TACC only drivers are the best drivers.

We're going to win!!

 

[Tam]

... highway data only...

Theoretically, just like GM Super Cruise, Autopilot can only be used on limited access freeways and not off-freeways (as warned by owner's manual.)

In reality, that is only true for GM Super Cruise but not for Autopilot.

GM Supercruise does not work off-freeway but Autopilot does work in many situations even in off-freeways.

Tesla was scolded by NHTSA for not geofencing enough and allowing Autopilot running during the Florida fatal accident because that highway although with a posted speed limit of 65MPH has intersections and is not a true limited access freeway.

Recent Consumer Reports prefes GM Super Cruise because it would be off for the above Tesla accident, off-highways, local roads, construction zones... It rewarded GM as the first place winner while Autopilot lost to GM even though it would still continue to work in those situations that the first place winner can't even work. (Consumer Reports thinks being conservative that does not work in many places means safe.)

So to answer the question: No. The data is NOT for highway data only. This is no first place winner GM Cruise. This is second place loser that works almost everywhere.

The criterion here is "Autopilot engaged" and it is used in almost all over the places.

 

[Tam]

...biased towards more complex and more dangerous situations...

No questions about it!

As it says in the manual, it "is primarily intended for driving on dry, straight roads, such as highways and freeways. It should not be used on city streets."

Currently, Autopilot takes the easy job and human takes the hard one.

But I think this is not about bias or unfair that who does a "straight" easy job and who does the more complex job.

Because without Autopilot, there's only one to take both the easy and hard jobs: human and there wouldn't be consideration about bias or fairness.

It's about: Is it safer to have your car with Autopilot where if you want, you can use it or a car with no options of using Autopilot at all (such as GM Supercruize.)

 

[caligula666]

No questions about it!

As it says in the manual, it "is primarily intended for driving on dry, straight roads, such as highways and freeways. It should not be used on city streets."

Currently, Autopilot takes the easy job and human takes the hard one.

But I think this is not about bias or unfair that who does a "straight" easy job and who does the more complex job.

Because without Autopilot, there's only one to take both the easy and hard jobs: human and there wouldn't be consideration about bias or fairness.

It's about: Is it safer to have your car with Autopilot where if you want, you can use it or a car with no options of using Autopilot at all (such as GM Supercruize.)

I get that but the point remains: there is no way of knowing that based on the statistic Tesla computed. It is a BS stat at best.

 

[Tam]

I get that but the point remains: there is no way of knowing that based on the statistic Tesla computed. It is a BS stat at best.

Before this, NHTSA cited the stat that Autopilot equipped VS. non-Autopilot equipped cars.

That is regardless whether an owner would use the Autopilot or not or even whether the owner bought the Autopilot or not.

So, now the criterion of "Autopilot engaged" is better than the above.

 

[strangecosmos]

Actually, NHTSA just looked at the crash rate for all Teslas before and after Autopilot was released. But the IIHS looked at Teslas with Autopilot hardware vs. those without and also found a reduced crash rate.