Early autopilot safety data. Tesla’s Autopilot lowers probability of having an accident by 50%

[RogerHScott]

You have to go a lot farther than that to really draw a valid scientific conclusion. What if there's a correlation between a driver's
intrinsic safety (behavior) and their likelihood of using AP? Even if the effect is real right now, what if much of it is due to heightened
driver attentiveness when using AP that will gradually decline as they become more comfortable with it and eventually fall to the point
where they're relying on AP too much and becoming overall less safe?

I really like AP, but I think it is reaching -- a lot -- to be making safety-related claims about it at this stage.

 

[ecarfan]

I don't want to turn this into fan boy battle but there is no science behind Elon's claim.

Please explain. Elon is obviously a pretty rational, scientific kind of guy.

 

[RogerHScott]

Please explain. Elon is obviously a pretty rational, scientific kind of guy.

I think "the science behind Elon's claim isn't very sound" is a more accurate way to put it. There is science, but not enough science and
not all of the right kind. I think any CEO who actually made scientifically sound statements to the public would be at risk of being charged
with neglect of fiduciary duty

 

[ecarfan]

I think if questioned Elon would admit that comparing AP miles driven to non-AP miles driven is not an apples-to-apples comparison. He has said many times that AP is for freeways and open highways only, and of course if weather conditions are bad (heavy rain, snow, fog) the AP may not be available.

Regardless, in my opinion it is still worth noting that the incidence of air bag deployment per X miles traveled is about 50% less on AP than when not on AP is an interesting data point. As AP becomes more capable and can be used in a wider variety of road conditions, I want to know if that difference holds up.

 

[mspohr]

I don't want to turn this into fan boy battle but there is no science behind Elon's claim.

I guess you do want to turn this into a fan boy battle.
Tesla has the data and Elon announced this as a preliminary finding.
We don't know anything about the analysis they did including whether or not they controlled for variables like road type, day/night, weather, etc.
So, any attack or defense of the statement is premature. Right now all we have to go on is an offhand remark at a meeting.
Rather than starting a flame war with the fan boys, I think it would be best to just wait until we have more information.

 

[AWDtsla]

I don't want to turn this into fan boy battle but there is no science behind Elon's claim.

Oh please this was probably half a line of SQL from Tesla's database. The sample size is quite large, and is growing rapidly at like a double exponential given sales of Model S/X.

 

[Dithermaster]

To the point above about near home (likely city) versus highway -- I agree! The correct numbers to compare would be highway miles (where Auto Pilot could be used but isn't) to highway miles (with Auto Pilot engaged). Then the numbers are real. Comparing all miles driven to Auto Pilot miles is not a fair apples-to-apples comparison.

 

[SageBrush]

What is the control group ?

 

[SageBrush]

To the point above about near home (likely city) versus highway -- I agree! The correct numbers to compare would be highway miles (where Auto Pilot could be used but isn't) to highway miles (with Auto Pilot engaged). Then the numbers are real. Comparing all miles driven to Auto Pilot miles is not a fair apples-to-apples comparison.

Very fair question. Or just a WAG -- the AP does not drive at excessive speeds ;-)

I also wonder if AP is only engaged during perceived good traffic conditions.

All in all, this is nowhere near a quality study so the results are interesting but should not be taken to mean much.

 

[EcoHeliGuy]

Two thing come to mind:

1) Elon roughly stated AP Miles vs non AP miles, 50% less likely to have an accident causing air bag deployment. He said "Autopilot" yet majority of the Tesla's on the road have the equipment standard for crash avoidance, so this is significant data.

2) Many commenters above are very confused. Statistics are evidence, not facts.