Another fatal autopilot crash - China

[somnambule]

How do you explain that the owner below had to manually brake to avoid collision to a regular height sedan in front as posted by @TRON ?

The car was traveling at about 48 to 50 MPH, the owner braked when it seemed not to automatically brake on its own.

If you look at the the dashboard, there is and icon of the car in front.

That means the radar did detect the stationary obstacle in front and reported it on the dashboard appropriately.

The driver in this case seems to brake quite early, likely before AP/TACC would react (esp. if the following distance was set to a low value). As you noted, this is not a case where the object in front was not detected as it's shown on the display.

I stand by my hypothesis that if AP/TACC couldn't detect a car stopped smack in the middle of the lane, there would be a lot more accidents being reported. I'm not aware of any cases where AP just plowed into the back of a stopped car in front. And I'm sure the opportunity has presented itself given the Joshua Browns of the world.

 

[3Victoria]

I think we need to wait for an official report. It is my understanding that the truck took the turn without stopping and therefore crossed quite quickly. The truck driver said that the car changed to the slower lane, and this suggests the driver was controlling the car, perhaps planning to go around the front of the truck, perhaps assuming the truck would stop. Instead the truck may have accelerated. There was also a suggestion of a rise in the road such that the two did not see each other until too late. Finally, the AP was designed to ignore objects stationary to the line of travel, as the truck appeared to the radar, so as not to brake for such items. Therefore the AP performed as designed (ie no fail).

Therefore, it is not clear what really happened and whether this accident was avoidable by Mr. Brown. Hopefully the report will clear this up ... And stop us speculating with poor information.

 

[brkaus]

Seriously, we have no reason to believe Mr. Brown would be alive if he had no AP? We have every reason to believe that based on the facts to date.

If I recall correctly, Joshua Brown was crediting autopilot with saving his life several months earlier? Or am I getting my stories confused?

 

[Canuck]

Therefore, it is not clear what really happened and whether this accident was avoidable by Mr. Brown.

Of course the accident was avoidable by Mr. Brown. That's not the issue. You must know that?

Hopefully the report will clear this up ... And stop us speculating with poor information.

This forum would be boring if we didn't speculate.

Plus, I believe when nikielizabeth posted what she did, from a friend of Mr. Brown, and I post what I do about AP1.0 lulling you into a false sense of security, and ohmman says "it's reasonable to assume that people pay less attention when on AP" we alert people, who may otherwise hear your explanations and take them as the possible cause such as: semis turn fast, rises in the road, and other non-AP related issues may be the cause, as if Mr. Brown was paying full attention while on AP but took no evasive action at all except changing lanes some time prior. To me, that makes no sense, but even if it does, and we are wrong, our message is to pay more attention, and don't rely too heavily on AP. Is that such a bad thing? Oh yes, it is, you gave me a "Dislike" for posting a heartwarming message from Mr. Brown's friends. Nice.

If I recall correctly, Joshua Brown was crediting autopilot with saving his life several months earlier? Or am I getting my stories confused?

There's a YouTube video he posted of his car taking evasive action and when Elon tweeted it out, he said he could now die and go to heaven. In watching the video, I don't think it saved his life, but it certainly looks to avoid a collision, and possibly could have saved his life.

But that's not the issue. I think we all agree AP does far more good than harm. The issue is not allowing it to lull you into a false sense of security, and making people acutely aware of the very little chance when it might do you harm.

 

[Gizmotoy]

Check the other two

The other two what? You posted one video. You claimed it proved the car didn't stop. I watched the video. The car stopped. The technology on the car in the 5 year old video is two generations old and doesn't have any applicability to recent vehicles. I don't know if you're being serious or just having a laugh at this point.

Edit:
Oh, I see. You went back and edited after I posted. You should be aware the the SUV didn't have tHe Technology Package needed for automated detection of pedestrians, as has been noted many times (it's even the top comment there with a link to an article about it).

The other video is even older than the first one you posted.

So, not very good evidence, no?

 

[S4WRXTTCS]

taking off a sweater?

You can't take off a sweater because the Tesla deactivates the AP when you unbuckle your seatbelt.

It's really awkward to remove a sweater with the seatbelt on.

 

[S4WRXTTCS]

If I recall correctly, Joshua Brown was crediting autopilot with saving his life several months earlier? Or am I getting my stories confused?

He credited it in preventing the accident, but his use of AP caused the situation to occur in the first place. Lots of people gave him flack for his lack of situational awareness in that near accident.

I'm not going to comment on his accident because he's the only one that really knows what happens. I think it's extremely tragic to die that way with people making all kinds of assumptions. Where he could have just had a sneezing fit or an eye lash that was bugging him.

 

[Canuck]

I'm not going to comment on his accident because he's the only one that really knows what happens. I think it's extremely tragic to die that way with people making all kinds of assumptions. Where he could have just had a sneezing fit or an eye lash that was bugging him.

If I'm ever killed in accident, I want people to speculate and analyze the hell out of it. If it saves just one life, I would be pleased (not really, since I would be dead). But I wouldn't want nonsense such an eyelash be used to dismiss a vigorous debate, especially when the debate may cause people to think "I better not do what he did" even if the speculation is wrong, but they think twice before using a laptop while on AP, even if I wasn't, and my eyelash killed me.

 

[S4WRXTTCS]

If I'm ever killed in accident, I want people to speculate and analyze the hell out of it. If it saves just one life, I would be pleased (not really, since I would be dead). But I wouldn't want nonsense such an eyelash be used to dismiss a vigorous debate, especially when the debate may cause people to think "I better not do what he did" even if the speculation is wrong, but they think twice before using a laptop while on AP, even if I wasn't, and the eyelash killed me.

I definitely don't want to die in vain, but you better believe I'm going to haunt you if you spread falsehoods.

To me it doesn't matter WHAT he was doing. We simply know that for some reason he was distracted or impaired. We know the AEB system didn't do it's job. That's enough without assuming anything about the driver.

His death is going to save hundreds of lives if not more as the result of the entire industry taking notice.

 

[Canuck]

I definitely don't want to die in vain, but you better believe I'm going to haunt you if you spread falsehoods.

No worries. I'm an atheist.

 

[S4WRXTTCS]

No worries. I'm an atheist.

I'm agnostic, and I hope that atheist are right.

I don't really want an afterlife. I like the idea of finality.

If I died and it wasn't over I'd be like WTF???? What now?

 

[3Victoria]

No, it is not clear that the accident was avoidable. I felt and feel you are reaching (and I have not seen that before.)

I agree, with AP one needs to pay attention.

 

[Tam]

The driver in this case seems to brake quite early...

Ok! Good theory but if I was that driver, I wouldn't wait to find out whether I brake too early or not.


Back to the Mercedes test:

1) 0-60 km/hr or 37.5 mph Drive Pilot stops by itself to avoid a rear end collision:

I think Tesla Autopilot can do the same.

2) 0-60 km/hr or 37.5 mph very noisy manual acceleration: Automatic Emergency Brake avoided a rear end collision:

I think Tesla Autopilot / AEB are not currently designed to interfere with manual acceleration. V7 dictates that manual inputs would override an automatic system.

I am not sure there is any plan to for Tesla to automatically brake when someone is stomping on the accelerator.

3) Lateral Turn Across Path (LTAP): Mercedes seems to handle this very well when there's a car running across its path.

For Tesla: As mentioned by MobilEye, it said currently the parts it currently provides to Tesla are not designed for Lateral Turn Across Path (LTAP). If you want that, you need to buy its newer version.

Elon Musk said Version 8 would be able to handle the Florida case which is what MobilEye called "Lateral Turn Across Path (LTAP)." And all that improvement without paying MobillEye for a newer version.

4) Braking for pedestrian: Mercedes seems to do it well at 35 km/hr or 22 mph:

For Tesla: There have been amateur tests by owners, to see whether AEB would brake for carton material (plain and aluminum foil wrapped) and even real human pedestrians. None reported any collision avoidance ability.

1) Bjørn Nyland


2) Ricco831: Youtube shows that the radar does recognize a carton box very very far away and does display on the instrument cluster as an icon of a car.

Repeated fails to brake to a stop to avoid hitting a carton box.

3) KmanAuto: Youtube shows repeated fails to brake to a stop to avoid hitting even on a real pedestrian.

I am not trying to be ungrateful of Tesla's hard work.

I am trying to be objective that Tesla's system is great but there is room for improvement and there is a need for owners to see a demonstration of its capabilities and shortfalls (rather than to find out what wouldn't work in an accident.)

 

[xav-]

If I recall correctly, Joshua Brown was crediting autopilot with saving his life several months earlier? Or am I getting my stories confused?

All on YouTube. I remember watching that video before his fatal accident was disclosed. Elon even tweeted about that save at the time

 

[S4WRXTTCS]

No, it is not clear that the accident was avoidable. I felt and feel you are reaching (and I have not seen that before.)

I agree, with AP one needs to pay attention.

What's frustrating about both this accident, and the Joshua Brown accident is blame is being redirected really out of fear. We fear that AP enabled the driver to be more distracted than they otherwise would have been.

The problem is in both cases there was a lot to blame.

In Florida there are a lot of fatality accidents on the road type the accident occurred on. It's an uncontrolled intersection on a high speed freeway. The semi also didn't have protecting guards to prevent a car from going underneath.

In this accident in China I find it completely unacceptable that there wasn't a truck with flashing lights behind the sweeper. It's on a freeway with speeds of around 70-80mph. This kind of accident seems like it would be common place, but we're going to blame AP?

These kinds of fatality accidents happen every day, and the only realistic way to stop them is through improvements to AEB systems in cars.

In both cases an improved AEB system like what's being introduced with 8.0 would prevent the fatality whether AP was on or not.

There was also a recent fatality accident in which the driver hit a tree and died. An improved AEB system might have even prevented the death of that individual, but in that case he was driving at a high rate of speed so it may have been too much for even a really advanced AEB system to slow the car down enough.

 

[somnambule]

Ok! Good theory but if I was that driver, I wouldn't wait to find out whether I brake too early or not.

Neither would I . But it does make this example useless for our discussion.

I think Tesla Autopilot / AEB are not currently designed to interfere with manual acceleration. V7 dictates that manual inputs would override an automatic system.

I am not sure there is any plan to for Tesla to automatically brake when someone is stomping on the accelerator.

I fail to see the distinction between AEB and TACC/AP if AEB doesn't override user input. The fact that Tesla talks about AEB separately from TACC/AP suggests to me that AEB is intended to override manual input. Now, that doesn't say how well it works; jus that it *should* work that way.

3) Lateral Turn Across Path (LTAP): Mercedes seems to handle this very well when there's car runs across its path.

For Tesla: As mentioned by MobilEye, it said currently the parts it currently provides to Tesla are not designed for Lateral Turn Across Path (LTAP). If you want that, you need to buy its newer version.

Here's a real live example of Tesla AP properly handling an LTAP case (and note that Tesla has disputed MobilEye's statement regarding this as they don't rely exclusively on the MibilEye system):

4) Braking for pedestrian: Mercedes seems to do it well at 35 km/hr or 22 mph:

As I noted in my prior post, I don't know if Tesla's system can detect pedestrians or not.

For Tesla: There have been amateur tests by owners, to see whether AEB would brake for carton material (plain and aluminum foil wrapped) and even real human pedestrians. None reported any collision avoidance ability.

1) Bjørn Nyland


2) Ricco831: Youtube shows that the radar does recognize a carton box very very far away and does display on the instrument cluster as an icon of a car.

Repeated fails to brake to a stop to avoid hitting a carton box.

3) KmanAuto: Youtube shows repeated fails to brake to a stop to avoid hitting even on a real pedestrian.

The problem I have with all of these is that none of them are performed within a properly marked lane (which the Mercedes test is). The fact that there's a known issue with identifying objects that partially obstruct a lane seems to suggest (at least to me) that the Tesla system is inherently tied to the notion of a lane. So the data we have from these videos isn't sufficient to say whether 1) the Tesla system would brake if there were proper lane markings and 2) the Mercedes system would work as well as suggested in the absence of lane markings. I'm not saying that tying AEB to a lane is the right thing to do. Just that we don't have the data to make the comparisons and draw the conclusions that you are suggesting.

I am not trying to be ungrateful of Tesla's hard work.

I am trying to be objective that Tesla's system is great but there is room for improvement and there is a need for owners to see a demonstration of its capabilities and shortfalls (rather than to find out what wouldn't work in an accident.)

Completely understood. Likewise, I'm not trying to defend Tesla. I do think there are corner cases where the system fails spectacularly (as we've seen a few times). It's just that I don't think there are obvious failures in simple, frequent situations. Otherwise we would be seeing a lot more AP-related accidents.

 

[Nick Deng]

There were mixed reasons:

1st to say was the cleaning truck should have other pre warning car behind like highway cleaning in CA, with much larger orange warning lights flashing and giving much distance for safty! But in China there isn't any! That is bad for public safety. As you can see from the video the shoulder is narrow only half size of a car.

2nd is the driver who was not paying much attention.

3rd, INDEED THE AP IS NOT PERFECT. I drove few times on the very 1st lane near divider like 605 fwy south to 5 fwy with narrow shoulder and AP is not good with concre block detection and line is not clearly detected either.

Sorry for the driver and lesson for all to learn use AP with the cautious!

 

[Tam]

Here's a real live example of Tesla AP properly handling an LTAP case (and note that Tesla has disputed MobilEye's statement regarding this as they don't rely exclusively on the MibilEye system)

Thanks for the youtube from Seattle Uber driver who got cut off by a car turning left from opposite direction in a rainy dark night.

Yes! That's a great example of "Lateral Turn Across Path (LTAP)!"

I guess I should listen to MobilEye with some reservations.

Good research!

 

[EVie'sDad]

Actually, we don't know if it is autopilot related or not. Tesla couldn't download the logs due to the severity of the wreck and the family apparently hasn't allowed direct access to the vehicle.

Also, the article I read mentioned it was the 23-year old son, who borrowed the car. So the theory that it was a complacent expert does not mesh with the greater possibility that this kid just trusted the tech too much and had not had enough education or experience how to use it, or blatantly chose to ignore the warnings. Any one of which could result in just this sort of disaster.

Condolences to the family, but the driver is responsible at all times for controlling their car, with or without TACC or AP enabled.

 

[Matias]

Anyone else equally concerned by the fact this happened in JANUARY? This occurred before the Florida crash and was not disclosed.

Additionally, all of the statistics that were previously stated may have been known to be false at the time they were said?

IMHO this is the most important thing in this accident. Elon never mentioned this accident in several occasions, when he discussed AP safety. Of course one can say, that it is not certain, that this was an AP related accident. But he still should have said "there is one fatal accident in China about which we are not sure whether AP was on".