Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

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

This site may earn commission on affiliate links.
Another streetview picture showing someone drifting to the left towards the gore area:

View attachment 290611

You know, showing that angle, I wouldn’t be surprised if AP locked on the outside line instead of the faded correct one. Also, this makes me wonder about the new wide lane support in 10.4. Maybe he had just got the update?
 
It depends on a lot of things, like how confident AP is, curvature of road, ... could be over a minute, I think.

That said, autopilot does not know if you are holding the wheel or not! Try holding the wheel but not resist any movements of the wheel in autosteer mode and you'll get the nag about placing your hands on the steering wheel.
Happens to me all the time, when I first bought my car I was squeezing the wheel at times strangling it thinking it was pressure sensitive. It is not.

I had to ask on TMC and people explained that it only senses an increase in torque meaning the weight of your hands resisting the action on the wheel. So if you hold the wheel and move your hands as it moves it won’t detect them at all. if you place a hand at the 10 o’clock position and let the full weight of your hand rest on the wheel then it will sense your hand

Another piece of advice. If you get off your ****ing cell phone and look at what’s coming you might be able to avoid a concrete barrier coming square at you. Just a thought
 
Just to let cooler heads prevail, i have had my model S for almost 3 years, and in that time it has hit the same pothole thousands of times. I always disengaged AP1 in this location and even submitted bug reports about it. Not once has AP ever changed its behavior, so why would I ever trust it to react differently? This was costly an error in judgment by the driver, sadly with very costly consequences. Despite his feedback, he chose to make a choice that ultimately cost him his life, but we who drive these advanced vehicles are intimately familiar with it's limitations and to expect otherwise is to tempt fate.

My condolences to Walter's friends and family.
 
This has been exactly my speculation. I personally have been fully convinced this was an AP accident (which says NOTHING about blame, the data just clearly pointed that way). I also was very curious after watching my car start to adapt on the commute to the wide lane feature, it was "trying" a bit too hard sometimes to assume the lane was wider when it should not. IN particular, when driving the other direction on 101 when the 2 carpool lanes merge into 1. It was too quick to try and "get in the middle".

Also, fully noticed the faded right side gore line, that is dangerous for AP.. as we all know from construction zones.

Very sad... I have definitely spent time on AP distracted, even for 6 seconds.. no more.


You know, showing that angle, I wouldn’t be surprised if AP locked on the outside line instead of the faded correct one. Also, this makes me wonder about the new wide lane support in 10.4. Maybe he had just got the update?
 
I find the visual cues/nags (flashing IC border) to be almost useless. My eyes should be on the road, not the IC (Eh, both periodically you say to avoid road hypnosis? Well, ok). And you only get so many subsequent audible cues before AP is disabled until you pull over, put the car in Park, and then resume. I vastly prefer how it *was* with Mobileye/AP1 when there was an audible chime and vibration - that was hard to miss, whereas this flashing IC stuff is too subtle to be all that useful all of the time. It's (too) easy now to tune out the current nags and then, all of a sudden not all that many miles down the road, you're without AP at all and are therefore arbitrarily in a less safe situation than you were in previously.

Mr. Huang was relatively new to AP (3-4 months if I recall). Am surprised he chose setting 1 of 7 for AP. By comparison, my default is 7 and sometimes 5, with collision warning set to early. Different commutes, different conditions, different drivers, I guess.

If I've comprehended the blog entry correctly, he last torqued the wheel at 6 seconds out, and the gore lane took 5 seconds to traverse.

It's unclear from the article whether Autosteer was engaged prior to or during those 6 seconds.

If TACC only was engaged, then it appears *he* drove into the end of the barrier by, for example, inadvertently following the solid white line(s), or swerving to avoid an encroaching vehicle.

When he torqued the wheel at 6 seconds, did he torque it hard enough to disengage Autosteer?
Was Autosteer engaged in the first place prior to the 6 second mark?
What was his collision warning set at/to/for (late, medium, early)?

And finally:
If Autosteer was engaged during those final 5-6 seconds, to what marking(s) was it tracking/aligned?

Taking a step back from this, it is... notable, for lack of a proper descriptor that does this situation justice... that a drunk driver *walked away* from impacting the smart cushion at 70mph in a lesser vehicle at this exact location ~2 weeks prior, and a noted engineer loses his life from impacting the unrestored area at we assume was about the same speed in the safest SUV ever made.
 
Looks to me the easiest explanation is he was driving along in AP, and it started tracking the wrong lines so it ended up in the gore section. It could have lost the tracking due to the sun. From previous comments, apparently there is a big change in dynamic range right around that area.

He had a good 5 seconds of visibility that he was in the wrong spot, and he didn't take action. So it seems like he wasn't paying attention for some reason. The only thing I find awfully confusing is he knew not to trust AP.

That's why I refused to believe it happened this way up to this point.
 
How does AP drive itself into the gore point on its own? Tesla has to know, right? Either theoretically or factually.

It’s difficult to accept this anomaly against all the other daily drives of the same area.

Current lane tracking is dependent on camera(s).

If I am a camera that see such a big lane, wouldn't I want to get in-lane too?

Picture from this thread:

img_1892-png.290277


Tesla knows!

It's imperfect and Tesla knows.

The easiest thing to program in this case is to forbid Autopilot to cross a solid white line.

Would you want Tesla to do that and restrict the freedom for drivers?

Autopilot will get better but I don't know when.

TeslaVision will classify objects and discriminate threats better.

HD mapping that Tesla has been collected will know that this is not a lane and it will prevent the car heading toward the gore point.

But all that is in future tense.

We are still at the present!
 
Last edited:
  • Like
Reactions: Sonic_78
IN particular, when driving the other direction on

101 when the 2 carpool lanes merge into 1. It was too quick to try and "get in the middle".

Oh, yeah, AP aggressively cut someone off on my commute home the other day because of that. It’s a lane merge where people routinely wait until the last. dang. second to figure out their lane is going away and merge. People also do last minute aggressive maneuvers to get around in the now shoulder (on the right side even).

As soon as the dashed line ended, AP 2 happily dove to the center of the new wider lane, nearly clipping the bumper of the car lollygagging in my blind spot waiting to merge.
 
  • Informative
Reactions: cwerdna
"In the US, there is one automotive fatality every 86 million miles across all vehicles from all manufacturers. For Tesla, there is one fatality, including known pedestrian fatalities, every 320 million miles in vehicles equipped with Autopilot hardware. If you are driving a Tesla equipped with Autopilot hardware, you are 3.7 times less likely to be involved in a fatal accident."

It is with certainty AP equipped Tesla will crash and lead to fatalities. When driver pays attention, AP provides extra level of safety. Data shows indeed most drivers pay attention, but unfortunately there are times both AP and driver fail to prevent an accident.

Yes, the above statistic is misleading. Teslas are bigger, stronger, newer and safer than most other cars. Their drivers are also richer. So it is natural that the fatality rate is lower. Everyone should own a Tesla!

However, the real question is: given identical Teslas, one with Autopilot and one without, which one are you more likely to die in?

Around the time of the last Autopilot brou-ha-ha, someone did an analysis here on TMC where they counted the number of AP fatalities (two) as well as the number of non-AP, non-suicide fatalities. Then they found public numbers for the number of miles driven on AP and miles not driven on AP, and concluded that you are something like 7x more likely to die in an AP Tesla as a non-AP Tesla. They just took the fatalities and divided by the number of miles, and made some assumptions to arrive at a SWAG that there was an 85% chance those numbers were not due to chance.

That counted fatalities, not collisions. The collision rate could well be different. And I'd be curious to see what the per-mile fatality rate is now for AP vs non-AP driving, though most people only activate AP on "safe" miles.
 
very deceptively crafted sentence. if i drove on AP for 15 minutes and i was alerted to put my hands on the wheel 2 minutes in.

It would fit tesla's specially crafted statement.

It makes it look like user was warned and didn't respond by putting their hands on the wheel but its completely false. The AP drove him straight into the barrier with no warning.
 
very deceptively crafted sentence. if i drove on AP for 15 minutes and i was alerted to put my hands on the wheel 2 minutes in.

It would fit tesla's specially crafted statement.

It makes it look like user was warned and didn't respond by putting their hands on the wheel but its completely false. The AP drove him straight into the barrier with no warning.

You're being warning to put your hand on the wheel explicitly for Tesla Motor's liability. What you should really be doing is paying attention. Where your hands are doesn't matter.