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

[Daniel Wilson]

View attachment 290972 View attachment 290973 This is how it’s done on the no speed limit Autobahn. It was introduced nationwide after a big discussion on guardrail slicing through passenger (car)s way back, I think in the seventies.

Fraction of the cost. Is the news full of airborne vehicles? No. Do they put a speed limit and signs out every time one is damaged? Yes.

Similar to the uk. No unmoveable wall

 

[Icer]

Left front rim is crushed, left front rim center section is down the 85 flyover lane, pack sustained impact near dead center. Likely cause: left tire hit left vertical member of barrier, pack hit right vertical.

The left front tire likely getting torn up when it hit the edge/end of the attenuator track.


OT:
Hoping for an informed discussion about how AP works.
Probably another thread, or a link to a good one would be helpful, thanks.

Main question is, that deadly spot had been passed a lot of times.
Why did Walter's car not "know" it is there?
*
Trying to imagine what the car "sees".
It's "Situational awareness"

Neural Network implied to me the Network is an active thing,
supplying info as part of the active AP process.

That lead me to conclude Tesla knew AP was on before checking the cars logs.
Now I don't know what I don't know. What had seemed obvious before is now
unknown.

 

[thegruf]

Did Tesla just slow down the roll out of the latest build?

I was expecting 2018.12 to be quite a wide release especially with Elon's new maps this weekend comment.
Not unusual in itself; but definitely seems to have slowed from what looked like intended wide release after only a couple of days.

And nothing wrong if they have, rechecking AP performance in light of this accident would only be prudent.
I for one would sacrifice a (possible) easter egg and new maps for a week or two in the greater good.

 

[AviP]

It means no inputs were detected, because:

1) he was distracted
2) incapacitated, such as a seizure or heart attack, or
3) hands on wheel loosely, below detection threshold

I imagine a full investigation would include the driver's medical and cellphone records, which of course we don't have.

The forensic analysis we've seen here is great, but unfortunately we also have Kia drivers suddenly becoming experts on AP behavior and playing armchair NTSB.

I'd like to throw in some more options because of the statement that he had 5 seconds and 150 m of unobstructed view and he had just turned AP on and had no input on the steering for 6 seconds.

4) He was video shooting the problem that he had complained about and did not realize how close he was to impact (objects in the iPhone are closer than they appear)

 

[thegruf]

Can I just observe that statements about the driver's actions/state of mind can only be purely speculative, and as such potentially hugely hurtful to the family at this sad time.

I really think it is best that this aspect of the discussion is restrained until a formal NTSB report is issued.

 

[Matias]

Problem with frequent OTA updates is, that AP’s behavior constantly changes. What is more alarming is, that those changes are not documented in the release notes.

This particular accident location has been traveled tenths of thousands times with AP-Teslas previously without problems. Now the newest praised version apparently has some specific glitch for this place. But there is no way a driver can know it, if the same car has previously handled that place without problems.

Tesla is playing a very dangerous game by constantly changing AP’s behavior without proper pre-testing and without documenting those changes to users.

 

[AviP]

Can I just observe that statements about the driver's actions/state of mind can only be purely speculative, and as such potentially hugely hurtful to the family at this sad time.

Point taken. I deleted the offending line.

 

[AviP]

Problem with frequent OTA updates is, that AP’s behavior constantly changes. What is more alarming is, that those changes are not documented in the release notes.

This particular accident location has been traveled tenths of thousands times with AP-Teslas previously without problems. Now the newest praised version apparently has some specific glitch for this place. But there is no way a driver can know it, if the same car has previously handled that place without problems.

Tesla is playing a very dangerous game by constantly changing AP’s behavior without proper pre-testing and without documenting those changes to users.

That is exactly what I was thinking all along. A lot of posters have been talking version numbers of the software. Realistically, even if Tesla documented everything perfectly and explained everything, who has the time to stay up-to-date with the reasoning and individual scenarios. Perhaps the AP afficionado, definitely not the regular customer. I don't know what the frequency of updates is, but if it is too frequent, then that means the software is Beta.

 

[timvracer]

This is a very good point. It is natural to start to "trust" AP when you drive the same route day after day, and it behaves exactly the same. It becomes very predictable...

until there is an update, and it is not.

Problem with frequent OTA updates is, that AP’s behavior constantly changes. What is more alarming is, that those changes are not documented in the release notes.

This particular accident location has been traveled tenths of thousands times with AP-Teslas previously without problems. Now the newest praised version apparently has some specific glitch for this place. But there is no way a driver can know it, if the same car has previously handled that place without problems.

Tesla is playing a very dangerous game by constantly changing AP’s behavior without proper pre-testing and without documenting those changes to users.

 

[JRP3]

This particular accident location has been traveled tenths of thousands times with AP-Teslas previously without problems.

Except for the fact that the driver involved mentioned problems in this exact area a number of times. He was well aware of potential issues in this very location, which makes this accident even more bizarre.

 

[JRP3]

I still don't understand why the radar system can't detect these stationary objects and at least try to slow down. Radar system doesn't have fast enough sampling rate at high speed driving? Not enough resolution to detect tiny object? Autopilot software not giving high enough priority for radar detection inputs? What's the deal here?

The barrier is a smaller cross section than even a small car so the system may ignore it to avoid false positives.

 

[mongo]

The left front tire likely getting torn up when it hit the edge/end of the attenuator track.

That could be, but I was talking about the rim/wheel which was crushed into an oval and the spoke section sheared off.

Edit: added "/wheel" to clarify.

 

[thegruf]

The barrier is a smaller cross section than even a small car so the system may ignore it to avoid false positives.

my train of thought also, together with the barrier being a planar sheet albeit slightly damaged, if it was even slightly not square to the car the return signal could have been very sketchy to the radar.

 

[ElectroCat]

The tesla had redundant steering systems, the organic system being the primary, auto pilot the secondary. It seems that a condition was found that both were incapable of handling, resulting in a accident. The serveity of the accident was the result of the condition of the safety barrier.
The organic based steering system is at he peek of its design with few improvements possible, while the auto pilot steering system is at the beginning of its design with many improvements possible, I'll vote for the system that can be improved.

 

[HyperMiler]

Would it be correct to expect that when AP has 10thousands of trajectories that are all superimposable on one another, that deviation from that path should trigger some alarm? What good is all the data collection otherwise?

 

[ricebucket]

Tesla's statement on this matter is disturbing. It's a whole lot of deflection that's more telling for what it *didn't* say. Namely, that most likely, Autopilot did not handle the split lane correctly and drove into the barrier.

Yes, the driver should have corrected before the collision, but if AP had done its job correctly, we wouldn't even be in this situation. Tesla didn't explicitly say this, but it's implicit in the blog post, blaming the victim for deficiencies in its own system.

I also find it odd that Tesla's first blog post said that Tesla will refrain from any statements until the conclusion of the investigation. But then promptly came out with a second blog post pointing fingers at the driver after the AP details were gathered.

 

[MXWing]

OK.



You just convinced me the driver was 100% responsible?

Driver was responsible 100% for the initial hit.

Driver likely would have survived if Caltrans maintained the barrier.

Two different things.

 

[mongo]

Tesla's statement on this matter is disturbing. It's a whole lot of deflection that's more telling for what it *didn't* say. Namely, that most likely, Autopilot did not handle the split lane correctly and drove into the barrier.

They said AP was on and that the car hit the barrier. What else is there for them to say?

Yes, the driver should have corrected before the collision, but if AP had done its job correctly, we wouldn't even be in this situation. Tesla didn't explicitly say this, but it's implicit in the blog post, blaming the victim for deficiencies in its own system.

AP is designed to follow lanes and/or other cars. The way the lines are painted, the gore point is a lane that ends in a barrier. EAP is not FSD.

I also find it odd that Tesla's first blog post said that Tesla will refrain from any statements until the conclusion of the investigation. But then promptly came out with a second blog post pointing fingers at the driver after the AP details were gathered.

Tesla downloaded the information from the AP computer, that likely concludes their investigation into the crash. They also did not "point fingers" they reported the data that was recorded: 150 meters unobstructed line of sight and no steering input.

 

[Ketchups]

The Mods rightly have requested this thread stays on topic re the accident on HWY-101, not a generalized discussion on lane split accident mitigation devices.

Attenuators are perfectly good and preferable to "launch type" rail ends as they contain an accident to the immediate area, reducing the probablity of secondary incidents; but only if they are in their operating condition. That the attenuator in this case was in a compressed state following a previous accident appears to be a key aspect of this sad fatality. Had it been in its correct operating state most probably this would have been a sever but non fatal accident.

yet as evidenced by this accident, a secondary and tertiary incident ensued.

I also wonder what eye-witness have said about the accident, as the mX could have been travelling the #1 lane perfectly fine and was simply rear-ended into the divider.

 

[mongo]

yet as evidenced by this accident, a secondary and tertiary incident ensued.

I also wonder what eye-witness have said about the accident, as the mX could have been travelling the #1 lane perfectly fine and was simply rear-ended into the divider.

Barrier was not reset. So car deflected off.
No one is reporting any other vehicle involvement precrash.

CHP report:

CHP Redwood City
CHP Redwood City
@CHP_RedwoodCity
·
Mar 23
Update on collision on US-101southbound at SR-85
Blue Tesla driving southbound on US-10, driving at freeway speeds on the gore point dividing the SR-85 carpool flyover and the carpool lane on US-101 southbound collided with the attenuator barrier and caught fire