Another Model X crash, driver says autopilot was engaged

[Garlan Garner]

All liars to the front of the class please. This person should have their picture posted on Teslas website.

Elon Musk on Twitter

Hey Consumer Reports.....where are you ?!!!

 

[Breezy]

I guess this brings up the question of where the cutoff is. So in this case, AP was disabled 11 seconds before the crash. In the van case, AP was probably disabled a second or two before the crash when he slammed on the brakes.

I think it depends on the circumstances. In this case I think 11 seconds is a suitably short period to suggest that Autopilot was likely a contributing factor. But if the driver had exited the highway, then crashed on surface streets 11 seconds later, then it probably wouldn't be AP-related.

When NHTSA says they are investigating whether driving aids were in use at the time, I don't think they're going to confine themselves to the moment of impact.

 

[Oceanwolf]

Agreed. I think there is still one improvement which could be made, though, which I'll talk about below.

Firstly, I think that job number one is to prevent a driver from becoming distracted, or incapacitated, in the first place. Once a driver has entered this state, there are much fewer options and things can go south quickly. In that respect, the driver failed. Autopilot's beeps and prods are designed to help prevent this, so in that respect, Autopilot also failed.

That's not really fair to Autopilot, though, since no matter what it does, the driver could still fall asleep, or worse. With level 2 autonomy there is little can be done, but still, there is something. I imagine this conversation between two Autopilot engineers:

A: "So, what do we do if the driver falls asleep?"

B: "We nag him awake."

A: "And if that fails?"

B: "We reduce power to the car."

A: "Great. But... you don't think he'd wake up in a panic, note that the car is slowing down, do something rash and roll the car, do you?"

B: "Well... maybe."

A: "But he deserved it."

B: "Yes."

A: "And what if someone is having a stroke?"

B: "Don't have a stroke."

To prevent this kind of situation, when disengaging, Autopilot should not allow driver input until the car has completely stopped. It's not like Autopilot was having trouble with the road. It could have continued driving, and thus would have been able to continue to stop safely. This prevents an incapacitated driver from immediately taking control of the car and doing something stupid when Autopilot decides to stop the car. After all, the driver is known to be incapacitated. They don't know exactly where they are, perhaps even what lane they are in, who is in their blind spot, how fast they are going, etc. And they may be groggy and paniced to boot. Why should they be allowed to resume driving immediately? They can do so after the car has stopped.

And then the car should lock all the doors, call the cops, give the location and report the driver.

 

[msnow]

None of it is broken

Beta can't be broken? Its not baked.

Hey! Whats wrong with this cake?
. It hasn't been in the oven yet.
I know, but its not the cake you promised.
. Didn't you hear me? I haven't baked yet.
I know but I ate a lot of it and it made me sick.
. Are you listening to me.
Yes, I know but .........

Of course it can be broken. I think you are trying to make a distinction between beta and production but from a functional perspective things can be broken.

 

[jdw]

K, based on the new info in the Japlopnik article, sounds like the driver fell asleep while in AP, woke up as the car was nagging him and then fell asleep again, woke up when the car started crashing. He was probably "sleep driving" when he initially took control of the car (ie. not fully awake), and probably won't remember that. That's my guess. Regardless, I do think we should not blame AP on this crash.

So, both the driver AND the passenger fell asleep?

"Albert Scaglione and his son-in-law Tim Yanke both survived the rollover crash last Friday near Bedford."

 

[Az_Rael]

OK, based on the new info in the Japlopnik article, sounds like the driver fell asleep while in AP, woke up as the car was nagging him and then fell asleep again, woke up when the car started crashing. He was probably "sleep driving" when he initially took control of the car (ie. not fully awake), and probably won't remember that. That's my guess. Regardless, I do think we should not blame AP on this crash.

The only thing I see potentially wrong with the sleeping theory is he had a passenger (his son in law). I guess both of them could have so asleep that neither noticed the warnings, but I doubt that.

 

[Garlan Garner]

Of course it can be broken. I think you are trying to make a distinction between beta and production but from a functional perspective things can be broken.

Of course. Yes things can not work properly, however if a vendor tells me "Hey, this widget you bought might not work as planned YET", then why would I expect it to?

 

[jdw]

So, both the driver AND the passenger fell asleep?

Quoting myself, but it gives new meaning to the old joke ..."When I die, I want to go peacefully in my sleep like my Father-in-Law, not screaming in terror like his passenger..."

One first class ticket please.

 

[Garlan Garner]

Quoting myself, but it gives new meaning to the old joke ..."When I die, I want to go peacefully in my sleep like my Father-in-Law, not screaming in terror like his passengers..."

One first class ticket please.

That was hilarious.

 

[vandacca]

Actually, when AutoPilot comes to a stop due to lack of driver input, that should be the last resort because it's a terrible option. Can you imagine the chaos that would ensue when a vehicle travelling 70 miles/hr on the freeway decides to come to a stop? I'm imagining a 15-car pile-up in such a situation.

The best option is for the driver to finally take over, and continue at speed.

 

[Garlan Garner]

Actually, when AutoPilot comes to a stop due to lack of driver input, that should be the last resort because it's a terrible option. Can you imagine the chaos that would ensue when a vehicle travelling 70 miles/hr on the freeway decides to come to a stop? I'm imagining a 15-car pile-up in such a situation.

The best option is for the driver to finally take over, and continue at speed.

If any car comes to a stop on the a freeway, then the driver needs their license taken away. Period.

 

[MP3Mike]

Actually, when AutoPilot comes to a stop due to lack of driver input, that should be the last resort because it's a terrible option. Can you imagine the chaos that would ensue when a vehicle travelling 70 miles/hr on the freeway decides to come to a stop? I'm imagining a 15-car pile-up in such a situation.

It doesn't just slam on the brakes. It turns the emergency flashers on and then starts to slowly decelerate. That shouldn't cause any kind of pile up.

 

[jdw]

It doesn't just slam on the brakes. It turns the emergency flashers on and then starts to slowly decelerate. That shouldn't cause any kind of pile up.

Stopping on a freeway is not ideal, but is probably less of a danger than bouncing off two meridians and rolling the vehicle.

 

[JohnSnowNW]

The only thing I see potentially wrong with the sleeping theory is he had a passenger (his son in law). I guess both of them could have so asleep that neither noticed the warnings, but I doubt that.

My wife routinely sleeps in the passenger seat, and is more or less dead to the world when doing so. So, it doesn't seem out of the realm of possibility...not that my wife's proclivities have any bearing in this specific instance.

 

[MP3Mike]

If any car comes to a stop on the a freeway, then the driver needs their license taken away. Period.

Really? I have had the fuel pump go out in my ICE while in the middle lane on the freeway, and because of traffic I couldn't get to the shoulder, my only option was to stop in the middle of the freeway, with cars flying by on both sides, and wait for the COMET truck to come and push me to the side and off the freeway. And amazingly there were no accidents as a result of all of that.

 

[Garlan Garner]

Really? I have had the fuel pump go out in my ICE while in the middle lane on the freeway, and because of traffic I couldn't get to the shoulder, my only option was to stop in the middle of the freeway, with cars flying by on both sides, and wait for the COMET truck to come and push me to the side and off the freeway. And amazingly there were no accidents as a result of all of that.

That's because you couldn't help it. Your car broke. You didn't volunteer to stop in the middle of the freeway.

 

[msnow]

Of course. Yes things can not work properly, however if a vendor tells me "Hey, this widget you bought might not work as planned YET", then why would I expect it to?

Wait, the post you quoted from me upthread was me saying not everything in AP is beta or broken which was a response to a question about which pieces of AP are in beta. Now you are trying to charactize me as against beta software. That just couldn't be farther from the truth. I paid over $100k because I enjoy this.

In looking at a few of your recent posts you (like our cars) seem to go from 0-60 pretty fast.

 

[vandacca]

It doesn't just slam on the brakes. It turns the emergency flashers on and then starts to slowly decelerate. That shouldn't cause any kind of pile up.

I realize that AP doesn't come to an immediate stop. The point is, the greater the difference in speeds between vehicles, the greater the chances of a serious accident. In this situation, 70 mph vs 0 mph is going to be pretty significant.

If all vehicles travelled at the same speed (e.g. 65 mph), then that's the most safest scenario. I've seen stopped vehicles "safely" on the shoulder get rear-ended. That's why in Canada, it's the law to provide service vehicles and police an extra lane when they're stopped on a highway shoulder.

In the AP case, I'm assuming it's going to slow down and stop in whatever lane it's currently in (not the shoulder) which is even more dangerous. One inattentive driver (not using AP him/herself) can plow into the back of the AP Tesla. Regardless, it's going to cause a serious traffic jam, if not worse.

 

[SR22pilot]

Fair enough. Although that will be very difficult to prove hardware/software failure since the driver is supposed to always be aware and able to take action.

I also consider the human factors effect of autopilot to be a part of autopilot, so I do consider autopilot related distracted driving to be something that may need to be looked into.

Do you hate cruise control too?

 

[SR22pilot]

I realize that AP doesn't come to an immediate stop. The point is, the greater the difference in speeds between vehicles, the greater the chances of a serious accident. In this situation, 70 mph vs 0 mph is going to be pretty significant.

If all vehicles travelled at the same speed (e.g. 65 mph), then that's the most safest scenario. I've seen stopped vehicles "safely" on the shoulder get rear-ended. That's why in Canada, it's the law to provide service vehicles and police an extra lane when they're stopped on a highway shoulder.

In the AP case, I'm assuming it's going to slow down and stop in whatever lane it's currently in (not the shoulder) which is even more dangerous. One inattentive driver (not using AP him/herself) can plow into the back of the AP Tesla. Regardless, it's going to cause a serious traffic jam, if not worse.

This is a heck of a lot safer than most other choices. My girlfriends father fell asleep at the wheel and dies. Had AP been on the car would have come to a stop and passengers would have noticed and woken him up. You don't want to brake hard to a stop and you don't want the car to go until the battery dies or the road runs out.