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

[barjohn]

I would like to clear up a misperception on the doppler radar. Doppler radar will pick up stationary objects. Doppler just means that part of the signal processing is to measure the frequency shift of the return to determine the speed of the object and since you are in a moving vehicle to subtract out the shift caused by your speed. In the Navy we used doppler radar for navigation all of the time. Other ship traffic in narrow shipping lanes and ability to see stationary channel markings in low visibility settings. Apparently, in Tesla's AP and probably other AP systems, the software is designed to filter out stationary objects to avoid false braking because there are numerous stationary objects that will be detected but are not a threat. This aspect of the software needs to be refined and improved. A window could be created with a limited vertical height from the road and a limited horizontal width slightly wider than the vehicle and any stationary object within a specified distance that falls in that window coupled with the vehicles known trajectory should start to apply the brakes to slow the vehicle down. With automotive radars having ranges up to 250 meters there should be time to determine, first there is a threat of collision and subsequently, the threat is not going away but in fact increasing in probability. The vehicle could be also alerting the driver. Where a sudden fixed obstacle appears like the fire truck scenario, it would have to take more aggressive action. If continuous awareness of the surrounding were part of the equation it could use evasive maneuvers if the maneuver would be safe to execute. More info on automotive radar attached.

 

[SMAlset]

I agree that there's an added reaction delay if your hands are off the wheel, but it's ultimately safer to be looking out the windshield with your hands off the wheel, than it is to be looking somewhere else with them on it. Especially since the wheel torque sensor is easily defeated. All the autopilot fatalities I've seen (perhaps even all the autopilot accidents, even), the Chinese instances, the semi, the fire truck, etc; could have been avoided if the driver had been looking forward. It doesn't matter what your reaction time is if you don't know you need to react.

Not arguing the need to be an observant driver but I'm very much a proponent of hands on the wheel. If you get a blowout (lots of crap on the roads these days--miss the days people didn't toss stuff out the windows and kept the road clean), you likely could be dead. Violent movement of your car to one side and even with both hands on the wheel hard to steer through. Easy to send car off to one side and take others out with you. And you'd want both hands on the wheel at that moment as well. As I recall this lesson was taught in my school's driver ed class on a simulator.

As for observing the road you don't to always just be watching ahead but periodically using your mirrors to view outvbach and to the side. Also easier on the eyes long term to have them not locked into one area.

 

[McRat]

...
WHAT I THINK CALIFORNIA SHOULD CHANGE
Duh... fix your damn roads, especially in a spot that is known to be dangerous. I personally put this fatality in CA's hands.

WHAT I THINK WE AS DRIVERS SHOULD CHANGE
Pay attention, use AP as it was designed to be used. It is not "self driving".

California has thousands of spots as or more dangerous than this one. All states do that I've driven, which roughly 40 including Alaska and Hawaii. The spot is only dangerous if you make the decision to ignore the road for 5 or more seconds. It is an obvious threat visible from a long distance away.

2018 cars are the safest in history by a wide margin. Our fleet of cars in the US is the safest it's ever been. Our fatality rate is rising.
As a nation, we do not WANT to pay attention to the road. A lot of it has to do with the change in cellphone tech which is now a device you must focus your vision on, not just listen/talk with.

But even with Listen/Talk only, we have found that your reaction times to emergency situation climbs dramatically when having phone conversion. Unlike talking to car passengers, a phone call has one party who cannot see traffic conditions, so does not delay speech or assist the driver to avoid threats. Quite the opposite. People don't really multitask like they think they do. Thinking about a conversation removes some of your decision making speed from the driving activity.

Things will probably get worse before they get better. Actual autonomy is the perhaps the fix needed. Which will be sad for us folk who like to drive, and the poor who cannot afford $70k cars.

 

[Krazaak]

Not arguing the need to be an observant driver but I'm very much a proponent of hands on the wheel. If you get a blowout (lots of crap on the roads these days--miss the days people didn't toss stuff out the windows and kept the road clean), you likely could be dead. Violent movement of your car to one side and even with both hands on the wheel hard to steer through. Easy to send car off to one side and take others out with you. And you'd want both hands on the wheel at that moment as well. As I recall this lesson was taught in my school's driver ed class on a simulator.

As for observing the road you don't to always just be watching ahead but periodically using your mirrors to view outvbach and to the side. Also easier on the eyes long term to have them not locked into one area.

I'm not sure it would have made much difference anyway, because whatever system is used to gauge driver attention is going to have a grace period. Obviously an eye tracking system can't require your gaze be locked forward 100% of the time and the current Tesla system doesn't require torque on the steering wheel 100% of the time.

Optimally, you'd be looking forward with your hands on the wheel the majority of the time looking sideways and rearward as needed. My issue with the Tesla system though, is that there seem to be many ways to defeat the steering sensor and hands on the wheel doesn't do much good if you don't know what's in front of you.

 

[alcibiades]

Autopilot doesn't "decide to bump into the barrier" & it isn't a case of "needing to wrestle back control" (per other posts).

When AP does depart a lane and head toward a barrier, is it that you don't like the word "decide" and thus you deny it happens? And when I do have to exert force to steer back, is it that you don't like the word "wrestle" and thus you deny it happens?

Are you saying those of us who have had these experiences are lying?

 

[kkboss]

Wow... go away for a bit and the board goes nuts.

It's not complicated folks... his hands were off the wheel for 6 seconds before the impact. Now each of you, look down at your phone and count to 6. Do you see how much time that is? Nothing extraordinary had to happen here, that is more than enough time for the car to follow the left gore marker (which was clear) instead of the right gore marker (which is very worn) in the morning glare, and simply drive directly into the barrier.

Now I know most of you are probably more responsible than I have been, but there have been many times I got caught in a distraction and took my eyes off the road for several seconds. Once even, the car did indeed start veering across my lane into a bus, but I sensed the motion and quickly looked up and grabbed the wheel. I fully knew I was doing wrong.. it was a dumb choice.

Tesla's incessant reminders to hold the wheel make it pretty darn clear that THEY WANT YOU TO BE ATTENTIVE. There is nothing as a 1 year driver that I could claim would make me think this car can drive itself.

And as I said in a previous post, I get it more now than ever, they mean it. No more multi-second distractions for me. I love AP, it has transformed my commute, and is indeed far safer than humans alone. I hate to see people trying to bring down this promising technology with a vastly better record for saving lives than cars without it for some personal agenda.

I feel terribly for this family, unfathomable loss. Unfortunately, we do have to compartmentalize here.

WHAT I THINK TESLA SHOULD CHANGE
I think when they release a new algorithm (in this case, wide lanes), they should require hands on for a calibration period for the driver to become familiar with potentially new behavior.

WHAT I THINK CALIFORNIA SHOULD CHANGE
Duh... fix your damn roads, especially in a spot that is known to be dangerous. I personally put this fatality in CA's hands.

WHAT I THINK WE AS DRIVERS SHOULD CHANGE
Pay attention, use AP as it was designed to be used. It is not "self driving".

Nobody has personal agenda here. We all agree to that AP technology need to be advanced but at what cost? Unfortunately, most drivers don’t understand the limitations as much as you do. They put trust in the AP because Tesla has advertised it in a way that drivers feel it can make your driving almost autonomous. Tesla should say that “AP can fail can cause death”. Because that’s what has happened. Hopefully, that will make people more attentive.

Moreover, on your 6 second logic, it absolutely is not enough time. If you drive on AP, you see how many times you get warnings. You get used to it. And then one real dangerous one happens, couple of second is already passed. You have 4 second to understand the situation and react to it. You need special training. Do not assume every driver to have the same level of motor skills and technology understanding as you do. Most of them buying the car assuming it will prevent such situations and not cause them.

 

[alcibiades]

It's not complicated folks... his hands were off the wheel for 6 seconds before the impact.

It's not complicated folks, that you don't detect something isn't proof that it isn't there.

 

[EVie'sDad]

Also I recently read a news article that approximately 49 percent of drivers have admitted to using a cell phone while driving. That is a staggering figure and if even close to being accurate, means many more of us are in danger on the highways everyday.

Please for you, your family's and friends and everyone else's around you sakes. PUT THE PHONE DOWN!!!

We don't know if this was a contributing factor yet, but if later analysis backs this possibility up, we will all have learned a valuable lesson about why not to do so.

 

[Krazaak]

When AP does depart a lane and head toward a barrier, is it that you don't like the word "decide" and thus you deny it happens? And when I do have to exert force to steer back, is it that you don't like the word "wrestle" and thus you deny it happens?

Are you saying those of us who have had these experiences are lying?

I would suggest that anyone who's claimed they had to wrestle control back from AP was exaggerating or there is something fundamentally wrong with the steering in their car. I've had AP disengage because it wanted to move 6 inches further to the right in the lane than my hands on the wheel would let it.

I'd be more willing to believe that with the hands off the wheel, autopilot could get a lot further into an incorrect maneuver and the driver would have to make a larger correction.

 

[robe900]

Nobody has personal agenda here. We all agree to that AP technology need to be advanced but at what cost? Unfortunately, most drivers don’t understand the limitations as much as you do. They put trust in the AP because Tesla has advertised it in a way that drivers feel it can make your driving almost autonomous. Tesla should say that “AP can fail can cause death”. Because that’s what has happened. Hopefully, that will make people more attentive.

Moreover, on your 6 second logic, it absolutely is not enough time. If you drive on AP, you see how many times you get warnings. You get used to it. And then one real dangerous one happens, couple of second is already passed. You have 4 second to understand the situation and react to it. You need special training. Do not assume every driver to have the same level of motor skills and technology understanding as you do. Most of them buying the car assuming it will prevent such situations and not cause them.

Agreed! Looks like new members are trying to enlighten older ones. Please see our point.

 

[alcibiades]

I would suggest that anyone who's claimed they had to wrestle control back from AP was exaggerating or there is something fundamentally wrong with the steering in their car. I've had AP disengage because it wanted to move 6 inches further to the right in the lane than my hands on the wheel would let it.

1. We all know that Tesla's power steering is great, so let's rule out that as a problem. Ba-dump bump. Ting!

2. When AP takes your car left, you have to turn it hard enough right to disengage. At speed, this results in a non-smooth change of direction and a physically non-trivial exertion of torque on the wheel -- especially if this is corrective taking of control and an avoidance maneuver. A trivial exertion of torque on the wheel is what Tesla uses to "detect hands".

 

[f-stop]

the only reason the distance setting of 1 (or 4 or 7) can be relevant is if a car in front suddenly changes to another lane leaving a corresponding time for the Tesla to re-acquire objects in front. I recall that Tesla specifically warn about this circumstance.

isn't this a possible contributing factor here?
in these now 60+ pages of this thread so far we here still don't have any firsthand knowledge of what the cars immediately around the MX were doing just before the accident. (At least investigators will have eyewitness accounts which we don't have).

Several members here that regularly commute on this route note that they frequently witness people cutting across this gore point at the last second. Others have also posted streetview examples of cars that appear to be driving directly down this same it's-not-a-lane for some distance. What if due to poor/faded lane markings and close distance setting (distance=1) the MX was closely following behind such a car, tracking it while AP (and driver) momentarily lost the correct lane markings? 70mph = ~100ft/sec, not a lot of time to react if not paying attention and the car in front dodges the very last second. For me at least, distance=1 is fine for TACC in city-traffic but too tight a margin for AP at highway speeds

 

[robe900]

isn't this a possible contributing factor here?
in these now 60+ pages of this thread so far we here still don't have any firsthand knowledge of what the cars immediately around the MX were doing just before the accident. (At least investigators will have eyewitness accounts which we don't have).

Several members here that regularly commute on this route note that they frequently witness people cutting across this gore point at the last second. Others have also posted streetview examples of cars that appear to be driving directly down this same it's-not-a-lane for some distance. What if due to poor/faded lane markings and close distance setting (distance=1) the MX was closely following behind such a car, tracking it while AP (and driver) momentarily lost the correct lane markings? 70mph = ~100ft/sec, not a lot of time to react if not paying attention and the car in front dodges the very last second. For me at least, distance=1 is fine for TACC in city-traffic but too tight a margin for AP at highway speeds

It was not following any car, there were no cars in visible distance in front of this car.

 

[Pando]

It looks like a confluence of many factors leading to the accident.

1. Road line markers worn
2. Different types of pavement (asphalt vs concrete) and the pavement gaps diverging differently from road line markers
3. No chevron markers in the gore lane to tell drivers they are not supposed to be there if they miss the diverging lines due to #1
4. No rumble strips to alert drivers they are driving in a gore lane
5. Distance from the lane divergence point to the barrier is way too short given the highway speeds in this area
6. Sun glare from the East at that time of day 9+am, plus the strobe effect from going behind and emerging from the freeway sign
7. Shadow from the freeway sign onto the road at crucial point where line markers start diverging, possibly confusing AP
8. Driver trusting AP more than he should
9. Car not equipped to detect stationary solid objects in the middle of the lane (not even alerting the driver)
10. Many hands-on-wheel alerts desensitizing the driver into false sense of security when there are no alerts (such as a scenario below)

So based on the conditions above, a speculative scenario unfolds:

A. Tesla follows a car into the gore lane
B. The car in front exits, realizing a mistake
C. Tesla continues straight ahead using the solid line markers on both sides in the gore lane
D. With this scenario (and I'm not saying that this took place), there is very little or no reaction time for the driver what is a few seconds ahead - a concrete barrier nearly impossible to see due to a car in front blocking visibility and sun's glare on the windshield
E. Crash barrier fully compressed, not effective, resulting in fatal injuries

 

[buttershrimp]

Ok, so I was wrong about AP not being active, I suck.

Now we need to modify the theory of what happened. We are now assuming that Walter thought he was in the adjacent HOV Lane? The one that would have taken him to the 101? I’m still confused as to how the car goes straight at that point. With a lead vehicle cutting him off an obscuring Lane lines for a moment? What’s the working theory now?

 

[FURY]

It has been many decades since my first motorcycle DMW test, but I seem to recall that there is somewhere in the California driving/riding motorcycle manual or recommendation literature that visual scanning every two seconds from side to side is what should be done. I may be wrong on that recollection and I don't know what the recommednatio, etc., if any is for four or more wheel vehicles, but ithat might be worth checking...

Thank you very much

FURY

 

[Krugerrand]

youre assuming the car was malfunctioning for 150 m, if it was driving by fine even after the driver saw the barrier but just bumped into the barrier, you cant say it’s not AP fault.
If it was equipment failure, it is the AP fault. If you’re car brakes don’t work, it is car’s fault!

No. If your brakes fail it’s just as likely your fault for not maintaining them.

 

[Barklikeadog]

Wow... go away for a bit and the board goes nuts.

It's not complicated folks... his hands were off the wheel for 6 seconds before the impact. Now each of you, look down at your phone and count to 6. Do you see how much time that is? Nothing extraordinary had to happen here, that is more than enough time for the car to follow the left gore marker (which was clear) instead of the right gore marker (which is very worn) in the morning glare, and simply drive directly into the barrier.

Now I know most of you are probably more responsible than I have been, but there have been many times I got caught in a distraction and took my eyes off the road for several seconds. Once even, the car did indeed start veering across my lane into a bus, but I sensed the motion and quickly looked up and grabbed the wheel. I fully knew I was doing wrong.. it was a dumb choice.

Tesla's incessant reminders to hold the wheel make it pretty darn clear that THEY WANT YOU TO BE ATTENTIVE. There is nothing as a 1 year driver that I could claim would make me think this car can drive itself.

And as I said in a previous post, I get it more now than ever, they mean it. No more multi-second distractions for me. I love AP, it has transformed my commute, and is indeed far safer than humans alone. I hate to see people trying to bring down this promising technology with a vastly better record for saving lives than cars without it for some personal agenda.

I feel terribly for this family, unfathomable loss. Unfortunately, we do have to compartmentalize here.

WHAT I THINK TESLA SHOULD CHANGE
I think when they release a new algorithm (in this case, wide lanes), they should require hands on for a calibration period for the driver to become familiar with potentially new behavior.

WHAT I THINK CALIFORNIA SHOULD CHANGE
Duh... fix your damn roads, especially in a spot that is known to be dangerous. I personally put this fatality in CA's hands.

WHAT I THINK WE AS DRIVERS SHOULD CHANGE
Pay attention, use AP as it was designed to be used. It is not "self driving".

There are two messages from Tesla.
1. Keep your hands on the wheel. You are fully responsible not us.

2. In 2017 the car will drive ny to la without the driver touching anything.
The new cars are FSD capable.
In 2019 you could sleep in your car while it drives you.
The car will uber for you...

 

[robe900]

What about recalls, still driver fault, huh ?

 

[mattack4000]

Yup apparently it takes super human skills to know how to drive.....

Btw this post is activating trolls in full force, love it.

It has been many decades since my first motorcycle DMW test, but I seem to recall that there is somewhere in the California driving/riding motorcycle manual or recommendation literature that visual scanning every two seconds from side to side is what should be done. I may be wrong on that recollection and I don't know what the recommednatio, etc., if any is for four or more wheel vehicles, but ithat might be worth checking...

Thank you very much

FURY