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

[NeverFollow]

This was my first though: The driver involved in the crash could have made a bad decision that day.

He might have decided to pass all the cars in front of him, by using the carpool lane on his left,
and then tried merging back on his right, but was not able to merge safely in time and hit the separation wall.

Could be, however Tesla stated that Autopilot was on during the crash and that the driver never intervened.
After reviewing all the pictures of the crash,
the location, and taking into account Tesla's statement based on the data recovered from the car
- Driver was likely not paying attention and
the Autopilot was following the solid left lane that led the vehicle straight into the divider.

I still don't understand how the car entered inside the gore lane.

My hypothesis would be that the driver was the one who initially
move the car toward the gore lane from left to right.

What really troubled me is this: If autopilot warned the driver at minus 5 second and 150 meters prior to the collision,
why autopilot not taking any evasive actions such as slowing down and or dis-engage the autopilot?
I would feel better if Tesla's autopilot at-least direct the vehicle to a safer environment.

Using Google map, I used the "50 ft" scale indication (see bottom right, red ruler) to estimate the length of the gore.
The total gore length is 600 ft or 200 m.

The distance from where the AP started to complain is 450 ft or 150 m.
The AP was complaining during 5 sec, so the car was driving 50 ft per second.
The estimated speed would be 108 km/h or 67 mph. (((150 / 5) * 60 ) * 60) = 108,000 m/h or 108 km/h

So what could have happen before the 5 second and 150 meters mark?

My hypothesis is that:

1. The car was on the very left lane BEFORE the beginning of the gore area with the AP activated.
2. Around 650 ft the driver touched the right turn signal to request the AP to change lane.

3. Around 600 ft the car started to move toward the right thus entering the gore lane.
4. Around 550 ft the car stop turning and stay in the gore lane.

5. Around 500 ft the driver should have realised that the car was in the wrong lane.
The driver should have keep pressing the turn signal to move forward on the right.

6. Around 450 ft or 150 m the AP noticed that something was wrong ?????
May be the AP already noticed that there was an obstacle 150 m away?
May be the AP noticed the trapezoidal shape of the gore lane?

7. During the last 150 m or 5 seconds, the AP was complaining (sound and flashing warning) but the driver didn't noticed.
May be the radio was on and he didn't heard the warning,
and the facing rising sun disturbed him and he didn't noticed the flashing warning...

8. May be the fact that the car distance was set to the minimum (one car or about 16 ft or 5 m)
the AP didn't slow down the car, because of possible false positive, and relied on the driver feedback to slow down.

 

[bebbiXpress]

I'd like to add something more general to this discussion that applies also to driving without any driver assistance features.

It seems as nearly nobody gives a damn about posted speed limits anymore. Going 5mph over the speed limit is kinda common behaviour and socially accepted. Here in this particular case is a 45mph limit posted well before that fatal divider! This tells me, the authority who did put that sign up thinks "that's a dangerous section"!

But in the now posted video on Youtube where someone tried to reconstruct how the accident may happened, the AP was set to 59mph and neither did the other cars you can see slow down to 45mph. Also from what Tesla stated about the accident (5 seconds and 150 meters of unobstructed view on divider) you can calculate that the crashed car was going around 67mph! That's more than 20mph over the speed limit.

What does that mean? Well, simple physics: 1st, the kinetic energy would have been half if he went 45mph. Just for comparison, hitting an immovable concrete divider with 45 is like hitting ground after falling from the 7th floor. Hitting it with 67mph is like falling from the 15th floor. 2nd, with 45mph there would have been additional 2.5 seconds time for the driver, EAP or AEB to maybe recognise the divider and react.

So, in the end it's your life and the lives of others you risk just to gain a few seconds on your way!

 

[22522]

I'd like to add something more general to this discussion that applies also to driving without any driver assistance features.

It seems as nearly nobody gives a damn about posted speed limits anymore. Going 5mph over the speed limit is kinda common behaviour and socially accepted. Here in this particular case is a 45mph limit posted well before that fatal divider! This tells me, the authority who did put that sign up thinks "that's a dangerous section"!

But in the now posted video on Youtube where someone tried to reconstruct how the accident may happened, the AP was set to 59mph and neither did the other cars you can see slow down to 45mph. Also from what Tesla stated about the accident (5 seconds and 150 meters of unobstructed view on divider) you can calculate that the crashed car was going around 67mph! That's more than 20mph over the speed limit.

What does that mean? Well, simple physics: 1st, the kinetic energy would have been half if he went 45mph. Just for comparison, hitting an immovable concrete divider with 45 is like hitting ground after falling from the 7th floor. Hitting it with 67mph is like falling from the 15th floor. 2nd, with 45mph there would have been additional 2.5 seconds time for the driver, EAP or AEB to maybe recognise the divider and react.

So, in the end it's your life and the lives of others you risk just to gain a few seconds on your way!

Are we sure it is not 55 or 65 there? In Texas it would be likely 65 mph.

 

[22522]

Here's another reason to be extra cautious with the new "wide-lane" support in AP2 on 2018.10.4: Lane correction when the vehicle is literally about to split two lanes at highway speeds is extremely quick. It felt more like an avoidance maneuver—as if it wanted to complete the lane change before the lane divider actually started on the pavement. I had both hands on the steering wheel (as usual) when driving, and it was still surprising how rapidly Autosteer moved into the right lane.

This happened at the lane split on I-280 South headed for the I-880 North/US-17 South exit, just past the Winchester Blvd exit (images are copyright Google from from Google Street View; they were not taken by me, and they are not from today):

View attachment 291377

The Google Street View vehicle is one lane to the left—the lane to its right is the one that splits into two lanes. With 2018.10.4, the wide-lane support kept my Model X centered between the ever widening "lane" until maybe 30-50 feet (very rough estimate) before the new lane divider started, at which point Autosteer moved rapidly into the right-hand lane (I don't know why the right lane was chosen vs. the left lane in this case):

View attachment 291378

I don't have a dash cam, and I don't care to have the publicity that might come with posting a video, but I want folks to be aware of this behavior. I'm considering reporting this to Tesla since it happens so quickly—it's basically the opposite of a smooth lane change.

Thank you for pointing this out. Lane centering does not match human behavior. And produces unnatural responses in a highway system designed around human behavior.

People sight the left lane line off the left front corner of the car to precisely position the vehicle with respect to that. The sight angles to the right lane line do not support alignment.

Peole do not do lane centering. They can’t.

They crank in more offset from the left lane marker, but they don’t center.

If the car wants to work well on roads designed for humans, it should act more like one.

Sorry for the rant. Listening function appears to be broken with the introduction of “wide lane” support.

 

[JTseng]

I still don't understand how the car entered inside the gore lane.

My hypothesis would be that the driver was the one who initially
move the car toward the gore lane from left to right.

Using Google map, I used the "50 ft" scale indication (see bottom right, red ruler) to estimate the length of the gore.
The total gore length is 600 ft or 200 m.

The distance from where the AP started to complain is 450 ft or 150 m.
The AP was complaining during 5 sec, so the car was driving 50 ft per second.
The estimated speed would be 108 km/h or 67 mph. (((150 / 5) * 60 ) * 60) = 108,000 m/h or 108 km/h

So what could have happen before the 5 second and 150 meters mark?

My hypothesis is that:

1. The car was on the very left lane BEFORE the beginning of the gore area with the AP activated.
2. Around 650 ft the driver touched the right turn signal to request the AP to change lane.

3. Around 600 ft the car started to move toward the right thus entering the gore lane.
4. Around 550 ft the car stop turning and stay in the gore lane.

5. Around 500 ft the driver should have realised that the car was in the wrong lane.
The driver should have keep pressing the turn signal to move forward on the right.

6. Around 450 ft or 150 m the AP noticed that something was wrong ?????
May be the AP already noticed that there was an obstacle 150 m away?
May be the AP noticed the trapezoidal shape of the gore lane?

7. During the last 150 m or 5 seconds, the AP was complaining (sound and flashing warning) but the driver didn't noticed.
May be the radio was on and he didn't heard the warning,
and the facing rising sun disturbed him and he didn't noticed the flashing warning...

8. May be the fact that the car distance was set to the minimum (one car or about 16 ft or 5 m)
the AP didn't slow down the car, because of possible false positive, and relied on the driver feedback to slow down.

View attachment 291385

View attachment 291383

Your hypothesis #7 can be valid if:
1. The AP's audio warning system failed
2. The driver is in panic mode and forgot to take control?
3. The steering's manual overide failed (driver took control but can't)
4. The real reason we have yet to uncover?

 

[Ketchups]

In the moments before the collision, which occurred at 9:27 a.m. on Friday, March 23rd, Autopilot was engaged with the adaptive cruise control follow-distance set to minimum. The driver had received several visual and one audible hands-on warnings earlier in the drive and the driver’s hands were not detected on the wheel for six seconds prior to the collision. The driver had about five seconds and 150 meters of unobstructed view of the concrete divider with the crushed crash attenuator, but the vehicle logs show that no action was taken.

@NeverFollow The mX could have simply kept the #2 HOV lane, then followed the more pronounced left line during the split at the 600ft mark in your photo, thus entering the gore lane. There is no mention of lane changes. Have you seen this video yet?

IMO Five Mississippi is more than enough time to solicit some sort of human reaction.

@Tam if you can't see, you shouldn't let the AP drive for you. I'd hate to see you end up on the news like this.

 

[yesup]

Everyone is using Pilot ... pro, auto, co, the name itself.
Manufacturers with the "pilot"ed systems
'Co-Pilot' (BMW),
Co-Pilot360 (Ford),
'Piloted Driving' (Audi),
'Speed Limit Pilot sub-function' (Mercedes-Benz)
AutoPilot (GM years ago)
Auto-Pilot (Chrysler/Imperial years ago)
'Piloted Drive' (Nissan), and
Pro-pilot (Nissan)

The problem is NOT the term "Pilot".
It is "AutoPilot".

Most people have heard of the term "AutoPilot" and have a preconceived expectation of what "AutoPilot" can do.
Other examples above like "Co-Pilot" or "Pro-Pilot" are terms that people have probably never heard of before - and would have no preconceived expectation of what functions it represents.
As for GM and Chrysler uses years ago, the technology at that time was too primitive to make it happen. As compared with nowadays when you see self driving cars news (Uber, Google...) frequently, and people would automatically associate the term "AutoPilot" with self driving cars.

The bottom line for me is, marketers in all industries come up with all sorts of "better than what's really there" monikers for their products, and Autopilot isn't as bad as many. .... A few on TMC are constantly saying Autopilot is a horrible name and Tesla is the antiChrist for using it. Forum rules prevent me from describing exactly what I think of that, but you're probably getting the idea.
It bugs me a lot when Tesla gets pilloried for stuff that every other competitor and every other industry does.

I am one of those who think "AutoPilot" is a bad name given its current limitation, and nothing more than a driver assist function.

For sure, it is a attention grabbing marketing term to encourage people to drop $5,000 to get it. Unfortunately, it also attracts news media attention.
I am very certain that all these news about Tesla crashes would not even be reported had Tesla labelled it as "driver assist" rather than "AutoPilot".

 

[Ketchups]

I am very certain that all these news about Tesla crashes would not even be reported had Tesla labelled it as "driver assist" rather than "AutoPilot".

I am certain every Tesla "driver assist" crash would still be shown in a negative light portrayed by the media. Think 'special interest'.

 

[bebbiXpress]

Are we sure it is not 55 or 65 there? In Texas it would be likely 65 mph.

Yes, we are. From that recreating video (link)...

 

[sinv]

This seems how it happened. The AP drove it into the barrier.

 

[NeverFollow]

One big difference between the AP and any Navigation Assistance tools I believe
is that the AP doesn't rely on itinerary directions.

When driving on freeway, each exit is announced at least 1/4 mile (400 m) in advance.
The name of coming next two or three exits are also mentioned above each exit name.
Those indications are also readable from about 1/4 mile away.

So there are in general plenty of time for a driver for been driving in the correct lane.

However, I found the signage a little bit complex.

The following sign, about 3/4 of mile away, shows that the exit for the Hwy 85 will be the following next exit.
So, if you want to go on Hwy 85, you need to prepare to be on the left lane.
This is the default case, there is no mention of carpool lane.

The following sign is reserved for the two carpool lanes.
The left lane is an exit toward the Hwy 85 and the right lane is the normal lane.

The following sign give the list of the coming exits.
The first one is the Hwy 85 exit.
The exit are presumed to be on the left.

This is the last sign before the exit.
The sign is located at the beginning of the gore area and about 150 m of the split wall.

In this particular case, a driver would have the choice:
- to exit on the left, when using the carpool lane,
- or on the right, if not using the carpool lane.

When passing this signal, it would be already too late for making any lane change when on the carpool lane.

The fact that the car passed over the gore lane was wrong.
Was this caused by the driver decision or by the AP misinterpreting the lane marks is not clear to me.

To have the AP relying both on the itinerary and also on the road marks
would be a good way for determining which lane to use.

I assume that autonomous driving would have to rely both on the GPS mapping and the road lane marks. .

 

[Ketchups]

This seems how it happened. The AP drove it into the barrier.

Case closed.

 

[bebbiXpress]

One big difference between the AP and any Navigation Assistance tools I believe
is that the AP doesn't rely on itinerary directions.

EAP doesn't use navigation data as of now. See my test here:

 

[Matias]

This seems how it happened. The AP drove it into the barrier.

That looks very bad for Tesla. No matter how you spin it, AP makes active steering input to hit the barrier. It is not that it did not avoid it. It actively steered towards it.

 

[bhzmark]

This seems how it happened. The AP drove it into the barrier.

That's almost certainly what happened. Caltrans needs to fix their crappy lane lines and Gore markings.

That looks very bad for Tesla. No matter how you spin it, AP makes active steering input to hit the barrier. It is not that it did not avoid it. It actively steered towards it.

No. It followed crappy lane line markings.

 

[NeverFollow]

@NeverFollow
The mX could have simply kept the #2 HOV lane,
then followed the more pronounced left line during the split at the 600ft mark in your photo,
thus entering the gore lane. There is no mention of lane changes.

If the driver had noticed an issue with this particular splitting lane area,
I wish he had made a recording of it or used a dash cam, since he seems
to have mention it before to other people.

There should be a way to warn any driver, using an app like Waze for example,
to store the GPS location, so any AP could be informed of the potential risk.

@NeverFollow

Have you seen this video yet?

In "this" video, there is not visible mark on the right side of the gore area.
I imagine that in the case of a foggy day, with AP or not, there could be potentially a lot of accidents.

 

[diamond.g]

This seems how it happened. The AP drove it into the barrier.

Case closed.

Are solid lines weighted higher than dashed lines? Seems like AP ignored the right lane markings in favor of the (wrong) solid line. Or is this a part of a don't get off the highway feature?

 

[mengwong]

So that’s 10 plus lives per year being saved by autosteer.

The theory of risk homeostasis suggests that with autosteer in the mix, people will drive more carelessly, with the result that approximately 10 more people will die per year to make up the difference.

Risk compensation - Wikipedia

 

[22522]

Yes, we are. From that recreating video (link)...

View attachment 291436

Ahh, that is an Indiana re-creation. I have not driven that stretch of road. In Texas that might indeed be 50 mph.

Thank you for orienting me.

[Texas is a little bit less hypocritical about posted speed limits compared to most states. Some here are posted at 85mph. Traffic safety wise, Texas seems to care more about people seeing success on the highway. I saw some signage at the I-30 lane zipper yesterday that was repetitive and persistent. LED lit red swing-away arrows that progressively moved traffic out of the lane. I think they were solar powered. Does Tesla make road signs? ]

 

[bro1999]

This seems how it happened. The AP drove it into the barrier.

That is a clear example (on the same road at the same exact location no less) that there is a huge problem with AP that needs to be remedied ASAP. The fact that the owner in that recreation video likely duplicated what happened to William Huang and that it could STILL happen to someone else as I type this is alarming. Even Uber ceased all AV testing after its AV was involved in that fatal accident. The fact Tesla has not deactivated (or even turned off AP functionality for that specific stretch of highway in CA) is mind boggling.