Automatic Emergency Braking Failure, the Movie

[jdw]

The names (Auto Pilot in specific) seem to convey much greater capability than is currently implemented so I'd argue that even best intentions of people reading the manuals & disclaimers will still be surprised/caught out/disappointed when the high tech car isn't as high tech as perhaps they had come to believe.

Auto pilot in an airplane does pretty much exactly what AutoPilot in a Tesla does - maintains heading, speed (and altitude). It does not autonomously fly the plane. I do realize the general perception of airplane auto pilot is that it replaces the pilot, but in fact, it only reduces the pilots work load by maintaining a set speed/heading/altitude - no matter what. The pilot is still responsible for flying the plane, and if the pilot sets an altitude and auto pilot flies the plane into the side of a mountain, that's on the pilot.

Had to say it ...not trying to pile on anyones misfortune.

 

[Msjulie]

@jdw I 100% agree the driver is responsible and OP said as much - I think both the posts that talk about Auto Pilot in reference to aircraft are perhaps inadvertently supporting my concern; controlled flight aircraft are pretty much assured nothing is going to fly into them or across their path at altitude and have a very different set of responsibilities than those implied by Auto Pilot in a car... that's all I'm saying..

cheers

 

[Paddy3101]

@Paddy3101 Are you a pilot? If so I'd expect you to know that distinction but I feel like the message to the general population is quite a different one.. I agree, the current software is clearly assistive in capability and I wish the 'marketing' of the feature was more open about that..

No, not a Pilot, but have worked on aeronautic navigation systems.

Looking critically at the video, and not placing any blame on the OP.

1) The OP fell asleep/dozed off/was not attentive. (Can't be for very long, unless there is a alertness defeat device fitted)
2) The car is driving at 65/70mph safely along the highway.
3) It didn't crash into the semi (reversing down the road at 70mph!)
4) It didn't veer out of it's lane and hit any other vehicles
5) It didn't veer out of it's lane and hit the guard rail.
6) Nobody was injured. (although road-workers could have been)

It's actually pretty impressive, and would have had a totally different outcome 5/10 years ago, and in virtually any other car on the road.

Marketing in one thing, and gets you in the showroom, its a sales tool.

Manuals and the information presented to the user in the car, when they sign-up for autopilot first and EVERY time you engage autopilot is another. EVERY time it tells you to keep your hands on the wheel and be attentive.

 

[Puddles]

Works great in the car but not so good in this forum . . . as you'll soon see.

Don't look at me. I've already noodled you about taking a face full of airbag-launched birdshot

 

[lolder]

The radar is currently the principal sensor for AEB. Barrels and cones may have different radar reflectivity based on many factors including composition and paint. The cars identify a "driving surface" by the cameras. The space between the barrels may be important for the cars to exclude the roadway on the other side of the barrels. People who are mostly water seem to have enough radar reflectivity to trigger the AEB. This will all come together with HW 3 and FSD software. There is currently no difference in the AP/EAP software performance on HW 2.5 or 3.0. No FSD software has presently been activated for general owners. I use EAP everywhere with alertness. It is getting better with each update even ones that don't list it in the notes. It now appears to resume speed after a lower speed limit in some cases. It also will allow more than 5 mph above the speed limit on SOME 4 lane, non-limited-access, divided highways. It takes sharper curves at higher speeds. FSD is going to happen soon.

 

[AlanSubie4Life]

Everybody is quick to attack the OP while also explaining away the limitations of the system as Should Have Known Better - it does have them, we all know that but why are we less vocal about the names chosen for these systems? Are the often exaggerated claims of what they are capable of not part of this problem?

That's one possibility for why this happens, but I'm not sure it's the primary reason. I have some ideas about other reasons why people might think (incorrectly) that the car will stop for pedestrians, barrels, cars, etc. But perhaps we should ask people here.

A few people have indicated here that they expect the car would stop more reliably for pedestrians. I have questions, along those lines, for people who have responded with that general opinion or other similar opinions (which may imply that they think the system has a reliable capability in some situations - or they may not think that...some people may have just been suggesting that the response to people MIGHT be more reliable than the response to barrels...not trying to put words in anyone's mouth here):

Why do you think the vehicle would stop for/avoid any obstacle? What is the thought process that goes into drawing that conclusion? What are the factors in your thought process? Is it the name of the assistance feature? Or is it something else? Is it your experience with the system? What about that experience gives you that confidence in the system?

Elon Musk has said that the most dangerous users of Autopilot tend to be the ones who have the most experience. So that's a clue about what is the flawed thought process that leads people to think the car will react to a given obstacle. But let's see what people here say...

 

[Saghost]

That's one possibility for why this happens, but I'm not sure it's the primary reason. I have some ideas about other reasons why people might think (incorrectly) that the car will stop for pedestrians, barrels, cars, etc. But perhaps we should ask people here.

A few people have indicated here that they expect the car would stop more reliably for pedestrians. I have questions, along those lines, for people who have responded with that general opinion or other similar opinions (which may imply that they think the system has a reliable capability in some situations - or they may not think that...some people may have just been suggesting that the response to people MIGHT be more reliable than the response to barrels...not trying to put words in anyone's mouth here):

Why do you think the vehicle would stop for/avoid any obstacle? What is the thought process that goes into drawing that conclusion? What are the factors in your thought process? Is it the name of the assistance feature? Or is it something else? Is it your experience with the system? What about that experience gives you that confidence in the system?

Elon Musk has said that the most dangerous users of Autopilot tend to be the ones who have the most experience. So that's a clue about what is the flawed thought process that leads people to think the car will react to a given obstacle. But let's see what people here say...

Mainly because at the Autonomy Day they talked about how they'd spent a whole lot of time making sure it recognised pedestrians.

Also because I've seen it draw them on the instrument cluster. I've never seen it draw a construction barrel.

 

[Magnets!]

Lucky you didn't kill a construction worker.

 

[AlanSubie4Life]

Mainly because at the Autonomy Day they talked about how they'd spent a whole lot of time making sure it recognised pedestrians.

Also because I've seen it draw them on the instrument cluster. I've never seen it draw a construction barrel.

Ok. But what makes you think it would stop for them, at least in a manner which is useful? Or do you think it would not reliably stop? I don't want to assume anything about what you think the system is actually capable of.

Just to describe my thoughts on this, I don't think the system will stop for anything, in a practical sense. I expect that I will have to stop for any obstacle, even when I am in Autosteer + TACC. I don't know whether or not you have a similar expectation, or a different expectation.

 

[lolder]

I am an experienced user of APs and cruise controls for over 60 years in cars, boats and jumbo jets ( L-1011). I'm an EE and retired airline captain. I am very aware of difficult AP challenges. The M3 handles most of them better than I expect. The two most common glitches are phantom braking which has become more prominent but less forceful and a tendency to take left turn lanes. The first is understandable as Tesla tries to mitigate the crossing tractor trailer risk. The second is still perplexing as sometimes the left turn wiggle is on well marked roads and yet the car will drive well on unmarked roads once EAP is engaged. I think the car is sometimes confused by a lot of bright lane markings. The left turn wiggles usually resolve but they disturb riders. When NoA is enabled on local streets the wiggles will probably disappear.

 

[Saghost]

Ok. But what makes you think it would stop for them, at least in a manner which is useful? Or do you think it would not reliably stop? I don't want to assume anything about what you think the system is actually capable of.

Just to describe my thoughts on this, I don't think the system will stop for anything, in a practical sense. I expect that I will have to stop for any obstacle, even when I am in Autosteer + TACC. I don't know whether or not you have a similar expectation, or a different expectation.

These days it sees stopped cars and stops for them, sees stopped motorcycles and stops for them. Both have to be basically camera based because the radar has trouble distinguishing different stopped signals.

I haven't had a case with a pedestrian in my path yet, but it shows me the security guard as I drive up, so my expectation is that it will stop for a pedestrian in the road.

I'm not dumb enough to assume it will or take chances to give it an opportunity - if the situation comes up I'll watch carefully and intervene if it looks close, but that's my expectation.

OTOH, I had a situation like the OP come up in NoA, and the car never plotted or reacted to the barrels, so when they started getting close I switched lanes manually (I made sure there was no one in the other lane near me before letting it develop to see what would happen.)

 

[Paddy3101]

A few people have indicated here that they expect the car would stop more reliably for pedestrians. I have questions, along those lines, for people who have responded with that general opinion or other similar opinions (which may imply that they think the system has a reliable capability in some situations - or they may not think that...some people may have just been suggesting that the response to people MIGHT be more reliable than the response to barrels...not trying to put words in anyone's mouth here):

Why do you think the vehicle would stop for/avoid any obstacle? What is the thought process that goes into drawing that conclusion? What are the factors in your thought process? Is it the name of the assistance feature? Or is it something else? Is it your experience with the system? What about that experience gives you that confidence in the system?

"Warning: Although Traffic-Aware Cruise Control is capable of detecting pedestrians and cyclists, never depend on Traffic-Aware Cruise Control to adequately slow Model 3 down for them. Always watch the road in front of you and be prepared to take corrective action at all times. Failure to do so can result in serious injury or death."

"Automatic Emergency Braking does not apply the brakes, or stops applying the brakes, when: • You turn the steering wheel sharply. • You press and release the brake pedal while Automatic Emergency Braking is applying the brakes. • You accelerate hard while Automatic Emergency Braking is applying the brakes. • The vehicle, motorcycle, bicycle, or pedestrian is no longer detected ahead."

"Warning: Several factors can affect the performance of Automatic Emergency Braking, causing either no braking or inappropriate or untimely braking, such as when a vehicle is partially in the path of travel or there is road debris. It is the driver’s responsibility to drive safely and remain in control of the vehicle at all times. Never depend on Automatic Emergency Braking to avoid or reduce the impact of a collision."

It's explicitly stated in the manual that it CAN detect pedestrians, using TACC and AEB.

It is explicitly stated in the manual that AEB will not deploy if it doesn't detect a vehicle, motorcycle, bicycle or pedestrian. (And it categorizes those, and puts them on the screen)

It is explicitly stated in the manual the Road Debris (my-words stuff that isn't categorized) can cause it to not brake for debris, or other things it should brake for.

Having said that, all of that is "IT CAN DO THIS" but its "NOT GUARANTEED" to do any of it.

The combination of "Real Driver + Autopilot" from the statistics does seem to be better than the real driver along. That DOES NOT mean that the Autopilot, on its own is better that a Real Driver.

It may well be, but there isn't any published data for that at the moment. All I can deduce is that the combination is better at reducing accidents.

 

[AlanSubie4Life]

Having said that, all of that is "IT CAN DO THIS" but its "NOT GUARANTEED" to do any of it.

Exactly. And that's a pretty big "but." I was just curious how many people have actually had the car slow for a pedestrian or bicycle, and are drawing the conclusion that the system is more capable than it actually is.

The combination of "Real Driver + Autopilot" from the statistics does seem to be better than the real driver along. That DOES NOT mean that the Autopilot, on its own is better that a Real Driver.

So far, the published statistics cannot be used to conclude that "Real Driver + Autopilot" is better than the "Real Driver Alone," for the same set of conditions. It's not knowable, from the datasets provided thus far. Frustrating, but true.

 

[AlanSubie4Life]

so my expectation is that it will stop for a pedestrian in the road.

In your mind, what probability do you assign to the car stopping for a pedestrian in the road? It sounds like you're watching out for it, so that's good. But curious about what level of certainty you have.

 

[Saghost]

In your mind, what probability do you assign to the car stopping for a pedestrian in the road? It sounds like you're watching out for it, so that's good. But curious about what level of certainty you have.

Right now, on a freeway? 80%? 85%?

It's all guesswork until we get some actual experience...

 

[alcibiades]

Right now, on a freeway? 80%? 85%?

It's all guesswork until we get some actual experience...

Any volunteers?

 

[AlanSubie4Life]

Right now, on a freeway? 80%? 85%?

It's all guesswork until we get some actual experience...

Interesting. At freeway speeds, I would assess the chance of stopping for a pedestrian at about 0%. I would assess the chances of slowing for said pedestrian at freeway speeds on a freeway at about 5%. Really just pulling numbers out of the air, to be honest - guesswork, as you say.

 

[DOCAL]

@Paddy3101 Are you a pilot? If so I'd expect you to know that distinction but I feel like the message to the general population is quite a different one.. I agree, the current software is clearly assistive in capability and I wish the 'marketing' of the feature was more open about that..

Constant rate climb straight into a stall, no problem.
Hold a heading/altitude straight into the side of a mountain, you bet.
Let you fly straight into the ground, absolutely.

The autopilot in most planes will happily kill you in many different ways if you're dumb enough to let it. The problem is that the public don't understand this, they think it's a magic box that will let the pilots essentially do nothing for the duration of the flight. So whilst autopilot is technically a great name for what the Tesla system does, there's a large gap between what it does and what the general public expect an autopilot to be able to do.

 

[Rockster]

So, the OP's position can be summarized by saying "Despite my being asleep at the wheel (literally, not figuratively), my Tesla stayed in its lane and suffered minimal damage from hitting plastic construction barrels instead of careening off of the guardrail or sideswiping an 18-wheeler. What could have been catastrophic loss of property and human life was instead a trivial amount of damage to my car's nosecone. And, I'm critical of this result."

And I wonder why my faith in humanity continues to plummet?

 

[Octo]

The disconnect between Tesla’s grand marketing for the future of auto pilot on tesla.com (“THE PERSON IN THE DRIVER’S SEAT IS ONLY THERE FOR LEGAL REASONS”) and reality of AP, EAP and FSD is awful.

Yes, blablabla million disclaimers that the real thing is actually *sugar* and what’s on the tesla site is SciFi / future only (for $6k today - buy now or else you’ll not be able to afford it !!!!11!1!!).

It has the potential for the *sugar* to hit the fan if they sell this to hundreds of thousands if not millions of people who believe the hype and don’t read the fine print. Just wait until some well connected lawyer’s family or a politician gets killed by someone who falls for Musks hyperbole and trusts his “Fully Self Driving” car. Good luck with that.