Model X Crash on US-101 (Mountain View, CA)

[22522]

The system only shows an idealized version of where it thinks the lines are.

It will also show you car icons (that look like Teslas) of where the car in front of you is, and maybe the next one further up the lane. (AP1 also differentiated between trucks and cars, and would show you vehicles ahead of you in the adjacent lanes.)

Thank you. If gore areas were added to its vocabulary and it painted them, I think folks would be a lot more comfortable.

 

[Doug_G]

Thank you. If gore areas were added to its vocabulary and it painted them, I think folks would be a lot more comfortable.

Although the line finding routines are obviously pretty powerful, I don't think the pattern recognition per se is anywhere near that sophisticated. It basically tries to find lines on either side of the car, and follows those.

 

[engle]

That's actually part of California VC, I believe. If a driver can't stop in time when the driver in front suddenly stops - or in your case, suddenly appears in front of you stopped - then the driver that couldn't stop w/o impacting the vehicle in front of them is "at fault".

That's why I really love the AEB on my AP 1.0 Model X P90D 000417, because it has radar, too. I also set my "driver warning of possible collision" (or whatever it is called), to "Early". I'd rather get false positives than miss a warning.

As a former, long time resident of the San Joaquin Valley, I am familiar with Tule Fog. (When 21 y.o., I had a minor collision in the fog going 25 mph on a road posted at 65 mph. Even though the vehicle I struck was making an illegal U-Turn in the middle of a busy highway and I hit it in the rear fender, the CHP said I was at fault because I must have been going too fast for the conditions if I couldn't stop in time!!!!!!!!! The other driver was cited for drunk driving.) Anyway, my question to you is, in the conditions you describe above, would you set your AP following distance to "minimum", or "maximum"? I know what I would have it set on in those conditions.

 

[T34ME]

Until we have a technology that can continuously monitor driver attention,.

Rear facing camera on the rear view mirror in the model 3? I bet we see that same camera in the MS/X interior refresh.

 

[T34ME]

I propose an autopilot kegel muscle sensor, one most have sufficient perineal muscle tone to enable autopilot.

As you age up, all those "safety" reflexes quickly dissipate. Believe me!

 

[engle]

One of my Model S neighbors told me a couple years ago one of the reasons he loves his Model S with AP is "now I can eat my breakfast in the car on the way to my office to save time in the morning". That uses both hands. I hope the crash victim wasn't doing any number of similar things while gazing out the side window or whatever before the crash...

[Personally, I keep my eyes glued to the road ahead while I'm distracting myself in my X -- normally shaving with an electric razor held in my left hand while my right arm is on my lap with hand loosely on the wheel so I can grab it almost instantaneously. I'm fully mentally-prepared to drop my razor, water bottle, or whatever in my hand if I ever need to.]

My only issue with AP 1.0 is indirect, since I've learned pretty much all of its foibles by now. On long stretches of clearly marked freeway, since I don't have to steer or control the pedals, all I have to do is watch the road. Sometimes, this boredom causes a little drowsiness. When that happens, I wait for an empty stretch of road, slow down a bit, grab the wheel with both hands, and push that metal pedal to the floor. The rush of acceleration adrenaline works every time! It's even more fun psychologically when I do it after changing my X to metric (160+ km/h). [yes, I have a radar detector.]

It works as designed, but the IMHO design is ineffective. You get warnings when you are holding the wheel normally. In order to prevent the warnings you have to provide some constant torque to the wheel - enough for it to register but not enough to cause the autopilot to turn off.

I used to use a different technique. I would give the wheel an occasional wiggle while driving, but found that this wasn't reliable. I'd get hands on wheel warnings periodically anyway, sometimes right after a wiggle. I think their sensor technique integrates torque over a period of time, so it is better to provide a constant small torque.

The first problem is that this is annoying. The second problem is it doesn't in any way provide an indication of whether I'm paying attention to the road, which is the main safety issue. Having your hands on the wheel, versus sitting on your lap, might make a difference of a few hundred milliseconds in response time. Not paying attention has a much, much larger effect.



I agree on both points. This is all about having some indication of driver engagement, but this is a somewhat intractable problem that exists even if the car doesn't have autopilot. I guess they're doing what they can, with what hardware they have in the car.

There is one thing in Tesla's response to the accident that I think is misleading:



Just because the "driver's hands were not detected on the wheel for six seconds" does NOT mean that his hands weren't on the wheel. As an example, he might have been using the "occasional wiggle" technique that I used to use. Tesla cannot claim that they know his hands weren't on the wheel. They don't know that. Which is the reason for the wording they used.

>>> agree 100% I keep at least one hand loosing on the wheel without torquing it until the X tells me too or I need to steer it.

Now if they mean to imply that the driver was not paying attention... well obviously he wasn't. That's not from the wheel torque sensing. For at least 5 seconds, and in all likelyhood far longer than that, he was not paying attention to the road. But they really can't claim that their logs are proof of this. The proof is that he drove into a stationary object.

 

[engle]

Agree if the 7/10 is really true, then continuing to use AP there is plain dumb. IIRC, before that little fruit company, the crash victim was a software developer at a video games dev shop. Hopefully, he wasn't addicted to playing in his X on AP 2.0 on his little fruit phone!?

PS. After years here, I think I finally figured out why so many of you folks have thousand+ posts. Quick, 1-2 liners?

So a married father took a revolver to his head and knowingly played Russian roulette at a minimum of 7-10 times? 'Rule of 3' could inflate this number even higher.

"Honey, you never guess what's been going on with Autopilot! It tries to kill me every single time I'm going to work. Thank goodness I always remember to turn it off and turn it back on everytime. Since Tesla didn't believe me the first 7-10 times it tried to kill me, I'm going to keep showing them."

Dunno WTF dinner conversation my wife would be having with me. Intentionally trying to make a orphan of my children lands me in divorce court pretty much immediately.

If playing Russian roulette wasn't interesting enough, he had to add playing on his phone at the same time. (Absolute, best explanation for 5 seconds of being in outer space).

 

[dgatwood]

Although the line finding routines are obviously pretty powerful, I don't think the pattern recognition per se is anywhere near that sophisticated. It basically tries to find lines on either side of the car, and follows those.

And if it finds too many lines, it can get very confused.

 

[buttershrimp]

Agree if the 7/10 is really true, then continuing to use AP there is plain dumb. IIRC, before that little fruit company, the crash victim was a software developer at a video games dev shop. Hopefully, he wasn't addicted to playing in his X on AP 2.0 on his little fruit phone!?

PS. After years here, I think I finally figured out why so many of you folks have thousand+ posts. Quick, 1-2 liners?

1-2 lines of Cocaine....

 

[Barklikeadog]

Agree if the 7/10 is really true, then continuing to use AP there is plain dumb. IIRC, before that little fruit company, the crash victim was a software developer at a video games dev shop. Hopefully, he wasn't addicted to playing in his X on AP 2.0 on his little fruit phone!?

PS. After years here, I think I finally figured out why so many of you folks have thousand+ posts. Quick, 1-2 liners?

Everyone keeps suggesting that he was distracted by his cell phone. We don't know that. What if in those crucial 5 seconds he was checking his mirrors? Maybe he was adjusting
something on the touchscreen. People recreating the accident were paying full attention and they were barely stopping in time.

 

[NeverFollow]

I am afraid the point is that Tesla Autopilot IS NOT SAFE

I had an accident January of this year.
Model S P90DL AP1
I was in motorway’s left lane and suddenly a truck went in my lane.
The system did not react at all!
Because of the very normal road conditions, normal speed, straight road, no rain I was relaxed and my intervention has been slower than normal.
I went into the truck and almost destroyed th front part of my Tesla.

Today I am not driving a Tesla anymore and I am writing this few words because I feel right to take part to this discussion.

The point is : ATTENTION

When AP is engaged every human being attention level will be lower than when you drive yourself!
I thougth much to what happened me and this point is very clear to me.
No way that attention level and reaction time can be equal to self driving!

So, until Tesla’s AP will improve to a reasonable safety functioning - and there is much work still to do - its use should be restricted much more than today.

Best Regards
Marco Merati

Does any Tesla model S/X/3 has a blind spot assistance warning?
In general, there is an orange light flashing near the side mirrors.

 

[SMAlset]

http://www.dot.ca.gov/research/rese...reliminary_investigation_12-12-13_revised.pdf

The south end of the 101/85 splits used to have one of those:
View attachment 291627
Are they out of "compliance" now because they don't have one of those at either end?

I wonder if they changed the regulations/guidelines recently to no longer require those "on the gore" signs... The study document above mentions wanting to just use overhead signs instead so that Caltrans workers don't have to spend time in the roadway repairing those "on gore point" signs.

But this brings up the point that it would be harder for a driver (and autopilot camera) to identify the actual gore point if they stopped putting risen, uniform signs on them like we see in that picture above.

I think maybe that bus crash broke the sign off, and they never replaced it...

View attachment 291628

View attachment 291630

In the case of the X crash, I think the only gore warning sign was very "close to the pavement", somewhat damaged, and the pavement markings are faded/worn out. There were also numerous old (no longer relevant) signs in the area that added to "visual clutter". Also questions if there were adequate lane delineators in the first place.

My sense is there is a tendency to favor road worker safety over driver safety as these things get redesigned. Sometimes you can help both, but sometimes there are trade-offs.

Pulled some photos from our dashcam from this morning to show the south end of 101/85 where the bus accident occurred as it appears today.



*Some of the lane lines look improved but not all.
*I did notice as you are approaching the divider/barrels, the left solid line on the 101 HOV lane looks completely absent...potential problem if you are in that lane??
*The 85 left lane flyover does start separating traffic a decent distance back from the divider wall.
*There are two sets of overhead signage indicating lanes that you pass under.
*The Diamond Exit sign that use to be on the divider wall never went back up after the bus crash.
*They did post an 85 HOV Ramp sign with a 45mph limit.
*I will say that the overhead signage that appears near the divider wall doesn't seem to be placed directly above the lane it corresponds to. Might just be the camera angle.
*Also just noticed that the overhead 101 signage doesn't indicate the most left lane is HOV which it is.

 

[Matias]

I don't know how you would define "compliance", but it looks like they repainted all the lane markings there, and went back to using barrels:
View attachment 291533

Again, this is a different 101/85 interchange. Down in San Jose where they rejoin together at the south end of 85.


It looks like the separator line for the "gone area" is black+yellow instead of a worn down white line.

But they still don't have any chevrons or plastic poles in the gore area to make it more obvious that nobody should drive there.

Maybe they want a whole lane there with no plasic poles so that the maintenance trucks can easily park there when they need to do repairs?


Also, some people are against using barrels because:
#1: They make a real mess when someone runs into them.
#2: They are harder (and more expensive) for Caltrans to replace than it is to reset a Smart Cushion.

I don't know anything about the relative driver safety between the different choices, but I will go out on a limb and say that I think the barrels are a lot safer than an already impacted / damaged Smart Cushion. They are also easier to see. They also seem far less likely to cause damage to a battery pack.

I think the lane markings and crash attenuator that were there before the bus crash were a lot harder to see.
View attachment 291534
It looks like that exit sign on the wall is no longer there. Maybe the bus tore it off and they never replaced it.

@smorgasbord ^^^

What those barrels contain?

 

[SMAlset]

Here's a how the HOV lanes divide - 85 flyover and 101 N. (greyhound bus accident area)

 

[Tam]

...following distance...

Good or bad, clear or foggy road conditions, I have always set my following distance the furthest, 7 since I first bought it a year ago, so if the system fails to brake or keep in lane, I can have time to react.

 

[power.saver]

*I did notice as you are approaching the divider/barrels, the left solid line on the 101 HOV lane looks completely absent...potential problem if you are in that lane??

The line to the left of the US 101 HOV lane is changing from white to yellow, which makes it difficult to see from the ramp. However, it should not change to yellow before the physical gore.

What those barrels contain?

Sand.

 

[Matias]

Although the line finding routines are obviously pretty powerful, I don't think the pattern recognition per se is anywhere near that sophisticated. It basically tries to find lines on either side of the car, and follows those.

If you look what Adrej Karpathy has done before, pattern recognizion to see gore areas should be easy for him. Of course he only has 24hrs in a day, but this should be number one priority (and it propably is now)

Andrej Karpathy Academic Website

 

[kkboss]

I did a little experiment today as I drive the same route every day. In manual mode, I drove my car and veered off the same way as its posted in one of the "recreated scenario" videos. I still haven't got my MX and now its part of the preparation to own it. I just wanted to see how much time I have to avoid that.

I realized how shrewd Tesla's overall statement is, especially the one about 5 seconds to act and 150 meters of unobstructed view.

I could see the barrier from more than 150 meters away, but as pointed out in the picture, the point where somebody could realize that the car is veering off and is going straight to the barrier(on AP) is not 70-80 meters away. That would mean at 70 mph, he had two seconds to react.
You got to be trained to avoid that.

In hindsight, this looks simple. If you drive the car expecting it to make mistakes and then you may be prepared to avoid this but most people drive the Tesla thinking the car has their back if they make a mistake.

Other statements from Tesla
Following distance 1: They have put this out on public domain probably as fact but irrelevant. Other than making the driver look reckless and irresponsible.

The driver had received several visual and one audible hands-on warning earlier in the drive and the driver’s hands were not detected on the wheel for six seconds prior to the collision:

The first part of the statement is to reinforce the perception that the driver was negligent.
The second part is where It raises doubts. If the MX was just following the left lane marking on the way to the barrier, the right side of the car would see more damage. From the pictures of the damage, it looks like the left side was hit hard. Did he try to steer back into the right lane to 101s?

There could be a delay in saving the data before the system would have shut down after the crash. Probably that's why it did not record the hands on the wheel.

After Tesla's statement in the public domain, it looks like they are worried about selling more cars and the stock price rather than making the roads safer.

 

[daktari]

"Honey, you never guess what's been going on with Autopilot! It tries to kill me every single time I'm going to work. Thank goodness I always remember to turn it off and turn it back on everytime. Since Tesla didn't believe me the first 7-10 times it tried to kill me, I'm going to keep showing them."

Are we sure that these 7-10 times it tried to kill him, or just a gentle swerve and back? Giving him a false sense of confidence, like a Styrofoam fence painted as steel. And then comes 10.4 and system behaviour changes...

In my view, AP maybe reduces some risks but also introduces new ones.

I also agree to those finding the wheel torque nag to be not good enough, a light grip is not enough. How many seconds from first nag to a warning sound? My previous Citröen tugged the seat belt as a lane departure warning.

 

[smac]

Can you cite any source that says any OEM is using lidar for steering or braking assistance on any passenger car?

The Audi A8 does.
https://spectrum.ieee.org/cars-that...st-production-car-to-achieve-level-3-autonomy


Now it will only operate up to a fairly slow 37 mph and on divided highways. This is it's defined ODD (Operational Design Domain).

As a level 3 car it MUST also have a "behavioural competency" to avoid static objects, whilst operating within it's ODD. This is the key difference between the Tesla and the Audi in it's classification of autonomy level.

Now I doubt many users have read the full specifications of SAE autonomy, or the federal guidelines, and while I have seen various comments by Tesla fans on articles of the Audi mocking it's slow maximum speed of operation, I really don't think all Tesla drivers understand this key difference. They are falsely assuming the Tesla has the competency (as have a number of commentators on this thread who assumed AEB would kick in), and as such is a level 3, not a level 2.

My prediction is automotive LIDAR costs will fall and resolution increase, alongside better processing allowing faster speeds defined in the ODD of such systems,. Subsequent to that a trickle down of the tech down from flagship models.

The checklist is summarised in this document, which is an interesting read: https://www.hsdl.org/?view&did=795644