Autopilot Warnings

[Potatoee]

What I find frustrating in these articles is the rather bland reporting about driver warnings prior to a crash. Example in this article includes "The report found that the driver’s hands were not on the wheel for the vast majority of that time despite receiving numerous alerts. Autopilot was engaged in the final 13 minutes and 48 seconds of the trip and the system detected driver-applied steering wheel torque for only 51 seconds of that time, the NTSB said."

Tesla Autopilot was engaged before 2018 California crash, NTSB finds – TechCrunch

I pretty much *always* have my hand on the wheel of my M3 yet I do get periodic warnings due to insufficient pressure. I would hate to get into an accident only to see a report stating that I didn't have my hands on the wheel when, in fact, I did. I'd like to get a "score" so I can challenge myself to do better. Perhaps this could be a feature in a future SW release (not sure if this is a serious suggestion on my part).

Alas we live in an imperfect world and we are responsible for controlling our vehicles but I do wish at times to see what the car thought I was doing and get better feedback.

Please no replies blaming autopilot for the above incident. I am a "I am responsible for my vehicle" person.

 

[GPinzone]

The Tesla system to check for driver's engagement is bad. Tugging on the wheel periodically does not mean the driver is paying attention to the road. I can't tell you how many times I get a "flashing blue" when my hand is on the wheel. The Model 3 has a camera pointed at the driver, but they don't use it. Why? They should have done what GM did with the Cadillac Super Cruise system and checked for the driver's eyes looking forward.

 

[Potatoee]

The Tesla system to check for driver's engagement is bad. Tugging on the wheel periodically does not mean the driver is paying attention to the road. I can't tell you how many times I get a "flashing blue" when my hand is on the wheel. The Model 3 has a camera pointed at the driver, but they don't use it. Why? They should have done what GM did with the Cadillac Super Cruise system and checked for the driver's eyes looking forward.

Exactly. I wish there was some rethinking of "how do we evaluate driver engagement" I think it's fine for now but could be better as we endeavor to improve the whole human/machine interaction architecture.

 

[Saghost]

They aren't showing us the right metrics IMHO. As noted above, the described time sensed on wheel metric can be a perfectly attentive driver with their hands needed elsewhere, or a driver who likes to leave their hand centered on the wheel, or someone who's barely paying attention.

More useful would be how long the driver took to respond to the prompts, how far along they got - any audibles?

I don't know for a fact that Tesla keeps a record of those aspects beyond the three strikes, but it seems like an obvious thing to do to me.

 

[PittCaleb]

I'm glad others found fault with the report. I used EAP exclusively on the highway (meaning I *never* drive on those roads). I always pay attention by any definition, but often my hands are either not touching the wheel or hanging on it which still requires some sort of periodic 'tug.'

The fault I have is that in order for the wheel to know I am active would require me to be attempting to steer it, no? I wholehearedly agree with the poster above, there needs to be another definition of 'driver engagement' - I believe one or more of the companies is using a driver facing camera and AI to determine this. Some combination of wheel engagement and cameras might be in order. And we all know there's a driver facing camera in our M3's.

I read that report and wondered what they would write about my hypothetical accident. "He was driving on I-80 for 100 miles and received 95 warnings to place his hands on the wheel. Airline pilots call that responding to the alarm."

That said,

• The longest period during which the system did not detect driver-applied steering wheel torque was 3 minutes and 41 seconds.

My longest would be what, 30 seconds'ish...

But... I would never run into the back of a firetruck at 45 mph because I'm paying attention. And I do use AP on more local roads (i.e. non-divided highways) even on my local-road commute. But I'm paying attention at all times.

 

[ewoodrick]

What I find frustrating in these articles is the rather bland reporting about driver warnings prior to a crash. Example in this article includes "The report found that the driver’s hands were not on the wheel for the vast majority of that time despite receiving numerous alerts. Autopilot was engaged in the final 13 minutes and 48 seconds of the trip and the system detected driver-applied steering wheel torque for only 51 seconds of that time, the NTSB said."

Tesla Autopilot was engaged before 2018 California crash, NTSB finds – TechCrunch

I pretty much *always* have my hand on the wheel of my M3 yet I do get periodic warnings due to insufficient pressure. I would hate to get into an accident only to see a report stating that I didn't have my hands on the wheel when, in fact, I did. I'd like to get a "score" so I can challenge myself to do better. Perhaps this could be a feature in a future SW release (not sure if this is a serious suggestion on my part).

Alas we live in an imperfect world and we are responsible for controlling our vehicles but I do wish at times to see what the car thought I was doing and get better feedback.

Please no replies blaming autopilot for the above incident. I am a "I am responsible for my vehicle" person.

So don't read a reporters interpretation of a NTSB report. They tend to add/change/forget a lot of words that have specific definitions. I'm pretty sure that there was something like "the car reported that the drivers hands were not on the wheels" a very big difference.
NTSB report summaries are very detailed and very specific. They just list the facts as they can be determined and then make a finding. And I think that in this case, they exonerated the autopilot, it was user error.
I believe that the interpretation of hands not on wheel for extended time actually led to the belief that the driver was completely inattentive.