Tesla's Autopilot needs to be shut down and NHTSA needs to do their due diligence

[DJ 240V]

With this issue hitting fever pitch (including blogs) , could we owners take a quick poll - Do you feel safe? [ Calling all HW2 EAP pilots.]

 

[GoTslaGo]

I'm friends with everyone, donchaknow?

Well, except for one "Keef Wivaneff". As I'd written above, this ex-Mercedes technician (9 years), who for decades has been one of the world's top dog mushers (Iditarod and Yukon Quest Champion, and multiple-time "Mushers' Choice" and "Veterinarians' Choice" award winner - i.e., a great sportsman) had never ever SEEN a Tesla before getting behind the wheel of mine....yet KW was the first to post a comment on the YouTube video Sab put up...an alarmingly scatological one...

Mercy.

KW also seems to be the only comment on the original LinkedIn article, linking to his own.... "wheels"...(in Teslabears...)

IIRC, KW had mentioned something about buying puts with the advice of some other Teslabears. Given the recent run up maybe prompting him to be more scatological to the extreme... bricks...

Well, OP now has the dubious honor of topping the "Best Posts" list for "disagree" ratings.

 

[S4WRXTTCS]

The video showing a car with AP1 hitting a traffic barrier demonstrates at least two potential problems with the current AP1 software.

The lane lines are ambiguous, which could be difficult for human drivers to determine where the lane is actually placed. The software didn't detect the temporary lane change - and proceeded straight ahead following the better marked lane.

The bigger issue is that the software did not detect the traffic barrier placed ahead in the car's current lane. There is always a risk that construction barriers, traffic cones or police will direct cars out of the defined lanes. AP/EAP are supposed to operate under driver control, so ultimately the driver is responsible for taking control and avoiding the unexpected barrier - though the software should be able to detect a large traffic barrier and do something reasonable.

FSD is designed to operate without any driver monitoring, even without anyone even in the car, if operating on the Tesla Network. That means FSD must be able to detect this situation and take actions at least as safe as a human driver would take.

Like with the accident when AP1 didn't detect the truck crossing in front of the AP car, Tesla should review this accident, and determine if the AP1 or AP2 sensor suite can detect this situation properly and the software can take reasonable actions. If the full AP2 sensor suite can't detect this situation, then Tesla may have a huge problem with EAP/FSD, because it's not that unusual to encounter temporary road or lane changes...

To me this demonstrates that we need road->car and car-car communication. So the construction crews can tell the road computer about the lane change. That way the car doesn't need to rely on visual clues as to where to go.

Even without AP that construction zone was dangerous. It came up so fast that if you'd crash if you were distracted at the wrong moment. Or if it was raining heavily at the time. Or someone in the other lane that didn't get over.

On my car with AP1 I would never ever use AP in a construction zone, and I really wish Tesla could use Waze or some other service to disable AP well before construction zones.

 

[Matias]

I don't know who in their right mind thought it was a good idea to start rolling out AP2 hardware before they even had basic feature parity with AP1, but those involved should be sacked.

I'm pretty sure it was Elon. Who else could it be?

 

[alevek]

And for every one time that happens it will have saved lives by avoiding approximately two or more human error caused accidents. And that's exactly why AP is safer.

Right. We'll see how that works out when an autopilot accident kills a carload of children coming back from a vacation to Disney World. You don't get credit for saving lives, but you do get lawsuits and investigations from accidents.

 

[brkaus]

My friends son's coach was killed this past Thursday night in an accident. The kids that were in his car with him driving back from the game are in the hospital, one in critical condition others in serious condition.

The oncoming car crossed the center lane line and hit head on. That driver also died.

The cause of the swerve is not yet identified. I won't even speculate, but neither car was a Tesla, so don't go blaming that.

The point is fatal accidents are a regular occurrence, now I know of one that hit home. Use of auto-pilot with hands on the wheel should be no less safe, and possibly safer as a momentary lapse on behalf of the driver might be corrected. But hands should be on the wheel, able to quickly detect if the car is heading out to the weeds.

As has been pointed out - Autonomous cars will never be 100% safe, sure. But neither are the folks currently operating heavy fast autos.

 

[calisnow]

With this issue hitting fever pitch (including blogs) , could we owners take a quick poll - Do you feel safe? [ Calling all HW2 EAP pilots.]

Nothing is hitting fever pitch except a few nerds on the internet hyperventilating.

 

[calisnow]

To me this demonstrates that we need road->car and car-car communication. So the construction crews can tell the road computer about the lane change. That way the car doesn't need to rely on visual clues as to where to go.

To me, all this incident demonstrates is that there are challenges ahead in self driving. As to what the actual solution is I don't pretend to know, but not being a computer scientist working inside Tesla's autopilot team it seems like a bit of a leap to conclude we know the best solution to avoiding barriers. If the neural networks are trained then what is so difficult about that situation?

Communication standards seem, to me, much more costly and difficult to implement than simply building smarter cars. Elon can use his brain to build smarter cars - but he can't use his brain to get the entire auto industry and government to coordinate and build a standard. In the mean time, what else should he do but charge ahead and keep making the cars smarter and smarter?

 

[whitex]

And for every one time that happens it will have saved lives by avoiding approximately two or more human error caused accidents. And that's exactly why AP is safer.

First where are you getting this 2:1 number for EAP? It hasn't even been deployed long enough to collect any statistically meaningful data. Second, how do you justify "save 2 people for the cost of 1" ? If you have 2 accident victims which urgently need transplant parts, how is it OK to kill and chop up one healthy person for parts to save 2 or more? I think Hitler had some beliefs like that. Do you?

 

[Robwoodruff]

I'm pretty sure it was Elon. Who else could it be?

I'm glad he made that decision. We had to wait 10 months with no autopilot after the hardware 1 was placed in the cars in 2014 but the software was not turned on. This time we had to wait 3 months before they decided to turn it on. Yes, it was very early, could only go 35 mph on the freeway, could not drive on local roads at all, etc. Every two weeks they send an update and it drives a little fast, and a little better. I would have hated waiting 10 months to have any autopilot in the new AP2 cars. Yes I wish AP2 could drive as fast as AP1 but I am patiently waiting for the next set of updates. The new cars are only using one camera at this time so I am hopeful that as the software is improved, it will be able to interpret the input from all the cameras, radar and sensors and drive much better.

 

[supratachophobia]

I am a former systems engineer, program and engineering manager for Lockheed Martin. There I worked on aircraft simulation, the Aegis Weapon System and was Software Engineering Manager for all of NORAD. I was also the whistleblower who raised the Deepwater Program issues - IEEE Xplore Full-Text PDF:

Uh oh, looks like someone just violated their NDA/security clearance..... The people I knew/know on Aegis, Phalanx, and Iron Dome, don't exactly go around telling people.

But back to autopilot. I can think of 3 instances in the last six months where APv1 tried to do something absolutely insane. And I have the dashcam footage to prove it. But I also had my hands on the wheel, which is where everyone else's should be so that these failures don't lead to the multitude of undesirable outcomes we've come to be made aware of. Now, I've had AP save my bacon around a dozen times in the same period, so how do we decide that the trade off is worth it?

 

[S4WRXTTCS]

To me, all this incident demonstrates is that there are challenges ahead in self driving. As to what the actual solution is I don't pretend to know, but not being a computer scientist working inside Tesla's autopilot team it seems like a bit of a leap to conclude we know the best solution to avoiding barriers. If the neural networks are trained then what is so difficult about that situation?

Communication standards seem, to me, much more costly and difficult to implement than simply building smarter cars. Elon can use his brain to build smarter cars - but he can't use his brain to get the entire auto industry and government to coordinate and build a standard. In the mean time, what else should he do but charge ahead and keep making the cars smarter and smarter?

First off neural nets aren't perfect even counting the human brain. There are definitely times when I'm driving that I simply do NOT immediately know where I'm supposed to go on some ridiculous roads. Not even counting really complicated round abouts.

Secondly neural nets can't do you any good if you can't feed them data.

You also need some level of redundancy in case one system fails. I've relied on other cars on the road numerous times in my life during bad situations. So why can't we give the car the same sort of capability?

Plus to really get efficiency gains autonomous cars should assemble themselves into trains. How can they do that without having a communication standard between the cars?

We already have communications standards for some many things. I can tweet from almost every corner of the world, but somehow my car can't communicate to the road?

There are people working on this and I can't wait for road->car, and car->car communication standards to arrive.

I see this as being more than just about autonomous driving. I see this as improving the situation for everyone on the road. Where even normal cars would at least get some kind of transponder to say "hey, I'm here" so even in adverse weather anyone with an active system would know a car is there.

Have you seen the dust storms that happen in Arizona?

 

[calisnow]

First off neural nets aren't perfect even counting the human brain. There are definitely times when I'm driving that I simply do NOT immediately know where I'm supposed to go on some ridiculous roads. Not even counting really complicated round abouts.

Secondly neural nets can't do you any good if you can't feed them data.

You also need some level of redundancy in case one system fails. I've relied on other cars on the road numerous times in my life during bad situations. So why can't we give the car the same sort of capability?

Plus to really get efficiency gains autonomous cars should assemble themselves into trains. How can they do that without having a communication standard between the cars?

We already have communications standards for some many things. I can tweet from almost every corner of the world, but somehow my car can't communicate to the road?

There are people working on this and I can't wait for road->car, and car->car communication standards to arrive.

I see this as being more than just about autonomous driving. I see this as improving the situation for everyone on the road. Where even normal cars would at least get some kind of transponder to say "hey, I'm here" so even in adverse weather anyone with an active system would know a car is there.

Have you seen the dust storms that happen in Arizona?

Obviously the more communication the better, but I'm not holding my breath waiting for standards. That particular example seems like a very simple self driving problem - one that Hardware 2 will easily handle once it is has a year or two of learning under its belt. As for things like dust storms - yeah I can see how comms would be quite useful.

Twitter is an example of something which has grown to be a defacto standard through sheer luck/timing - it never had to overcome a coordination problem from day one. Now, maybe some comm standard will take hold organically in cars too - that would be awesome if it does.

 

[Tangible]

The Greek prefix "auto" means "self". Merriam-Webster defines autopilot as "a device that steers a ship, aircraft, or spacecraft in place of a person". In the public mind, if not in an actual aircraft, autopilot is completely autonomous.

Giving that name to Tesla's feature set was and continues to be a source of possibly lethal misunderstanding and certainly disappointment. My other cars refer to their (rough) equivalent as Lane-Keeping Assistance. If Tesla wants to claim superiority they could have called it Super Lane-Keeping Assistance.

I know that there are numerous warnings about how to use the system and its limitations. These satisfy legal requirements but not the realities of human behavior. We've all been conditioned to skip the first few pages of any instruction manual where the legal stuff lives, and to "Agree" to 100-page software licenses that no one has ever read.

 

[stopcrazypp]

The Greek prefix "auto" means "self". Merriam-Webster defines autopilot as "a device that steers a ship, aircraft, or spacecraft in place of a person". In the public mind, if not in an actual aircraft, autopilot is completely autonomous.

Giving that name to Tesla's feature set was and continues to be a source of possibly lethal misunderstanding and certainly disappointment. My other cars refer to their (rough) equivalent as Lane-Keeping Assistance. If Tesla wants to claim superiority they could have called it Super Lane-Keeping Assistance.

I know that there are numerous warnings about how to use the system and its limitations. These satisfy legal requirements but not the realities of human behavior. We've all been conditioned to skip the first few pages of any instruction manual where the legal stuff lives, and to "Agree" to 100-page software licenses that no one has ever read.

Autopilot "steers" in place of a person but does not absolve the pilot from paying attention and being ready to take over (similar to how cruise control accelerates/brakes in place of a person). I think the name is a red herring, as even using the system for one time and having it take over once, already drives home the point the system can't be left to its own devices, warning messages aside.

I would also say that generic names like "Lane-Keeping Assistance" are not use to describe whole systems. They use similar marketing names like "Drive Pilot" (Mercedes), "Pilot Assist" (Volvo), "Active Driving Assistant Plus" (BMW).

If you think other automakers are above using marketing tactics that suggest that their systems offer completely autonomous driving, then you would be wrong (here's some examples of them calling their system "self-driving", something Tesla has carefully avoided).

Here's Volvo calling their system "self-driving":
Volvo Car USA on Twitter

Mercedes did this multiple times (both in print and video ads):
http://jalopnik.com/possible-mercedes-self-driven-advertisement-doesnt-feat-1783761678
http://jalopnik.com/federal-trade-commission-asked-to-investigate-mercedes-1784419463

 

[Tangible]

"If you think other automakers are above using marketing tactics that suggest that their systems offer completely autonomous driving, ..."

I do not. But I don't find "All the other kids are doing it" to be a compelling excuse for Tesla.

Good point about AP being the name for an entire system. That's no reason not to give it a more modest name. In fact the best name would be completely meaningless, e.g., Teslatronic. Teslariffic. Tess. Oh, I really like that last one. Who among us hasn't given a human name to a gps or a Roomba?

Note that automatic transmissions require no intervention, nor do automatic chokes. I don't need to stay up all night with my self-cleaning oven. All these things give rise to inferences.

My last quibble with Autopilot: I keep thinking of the inflatable guy from Airplane.

 

[bhzmark]

First where are you getting this 2:1 number for EAP? It hasn't even been deployed long enough to collect any statistically meaningful data. Second, how do you justify "save 2 people for the cost of 1" ? If you have 2 accident victims which urgently need transplant parts, how is it OK to kill and chop up one healthy person for parts to save 2 or more? I think Hitler had some beliefs like that. Do you?

NHTSA report regarding AP1.

Some people die because they are strangled by their seat belt, or their airbag hits them the wrong way, or their vehicle stability control prevents them from doing a certain evasive maneuver. But on avg those technologies save many more lives than they take. People who have passed their freshman year philosphy/ethics course have no problem understanding that. Keep studying up on the trolley car problem. This isn't one of those harder problems.

 

[pedriscoll]

I do note that the OP has not posted anything since the day he started this thread.

 

[whitex]

NHTSA report regarding AP1.

And what made you think that you can base the safety of EAP today on historical data of AP1? Are you saying EAP today is just as safe or safer than AP1? If not, don't you think it would have been safer to keep shipping AP1 until EAP is at least on par with AP1? Anyone that could have been saved by AP1 Emergency Braking is out of luck driving an AP2 car today.

Some people die because they are strangled by their seat belt, or their airbag hits them the wrong way, or their vehicle stability control prevents them from doing a certain evasive maneuver. But on avg those technologies save many more lives than they take. People who have passed their freshman year philosphy/ethics course have no problem understanding that. Keep studying up on the trolley car problem. This isn't one of those harder problems.

Logic behind seatbelts and such is different - there is a clear "benefit in most situations" argument. What Tesla is doing here is akin to a drug company which has a successfull drug, then coming up with a brand new forumulation , and using the success rate of the old drug to justify why they sell the new drug to people, watching the results and tweaking the formula. Would you advocate for such drug trials on the public and justify it with the greater good (which I'm sure you could get numbers for, since it definitely speeds up the drug development process to be able to test it on the general public).

 

[stopcrazypp]

"If you think other automakers are above using marketing tactics that suggest that their systems offer completely autonomous driving, ..."

I do not. But I don't find "All the other kids are doing it" to be a compelling excuse for Tesla.

Good point about AP being the name for an entire system. That's no reason not to give it a more modest name. In fact the best name would be completely meaningless, e.g., Teslatronic. Teslariffic. Tess. Oh, I really like that last one. Who among us hasn't given a human name to a gps or a Roomba?

Note that automatic transmissions require no intervention, nor do automatic chokes. I don't need to stay up all night with my self-cleaning oven. All these things give rise to inferences.

My last quibble with Autopilot: I keep thinking of the inflatable guy from Airplane.

Personally, I think Autopilot is a perfect name and I understood it as soon as it was used: like a plane's autopilot it steers the vehicle, but the pilot is still monitoring the situation. That's the reason Tesla gives for naming the feature that way.

Of course, some people like you think of the movie Airplane. However, even so, after using the feature the first few times, can you seriously say you feel the feature is fully autonomous driving that requires no driver attention? I simply do not believe any significant amount of owners really would feel that way and that the name is just a red herring.

More important is what mitigation strategies that Tesla used to get people to pay attention. And the steering wheel nags and strike out system is what Tesla uses (which some people dislike, but NHTSA feels it is necessary and sufficient).