rush6410
Member
3. He engaged autopilot while in the middle of the merging lanes which leads to the collision.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Model X Crash on US-101 (Mountain View, CA)
- Thread starter mookhead
- Start date
Yes, Tesla couldn't have screwed up these two things more badly than they have:
1 - (FACT) Managing the relationship with China. You never ever ever call China out in tweets. You discuss this behind closed doors so no one loses any face and both sides accomplish their goals.
2 - (MY STRONG OPINION) Spend $5 and put a speedometer in front of the driver on the Model 3. I am sure people can survive without it but its unnecessary distraction to not have it for such a small investment.
NOW... on the topic of this crash.
Unless Tesla prevented the driver from taking control of the vehicle, there really isn't anything to have to hold Tesla accountable for.
There just isn't.
If one wanted to question Tesla "should have tested AP 2018.10, 2018.12, etc more" - now that is a slippery slope.
I can't say much on this but I am being truthful that there are individuals with access to more advanced software builds than what the general population receives. Tesla does their testing and they have a team that helps them test.
At what point is the software "foolproof"? The answer is never and it's always a moving target.
You need to pick between
1.) Foolproof software (this is not a real option)
2.) Work in progress released in good faith after testing internally and externally.
Last edited:
I share this frustration.
The vast majority of companies communicate not at all.
Elon's responses on twitter are a dramatic shift from that,
very much appreciated and really do make
the whole thing feel different.
Improved.
But they are still hip deep in the muck, it is a war, a fight for
survival, lawyers and accountants and complexity and pressure
...I am thankful is not my problem
So I tell myself to be happy they are responding.
Even if sometimes it seems to be through indecipherable filters.
Work is in progress.
Keep the faith.
Perhaps 1 more way (and it does happen to me on occasion): 4) holding the wheel to keep the car travelling straight. If auto-steer suddenly tries to veer it'll be disengaged pretty much immediately as well.
That is advanced AP driving technique that actual owners might not know about.
Explaining this to people who don't own Tesla's and have no AP experience is describing color to a blind person (as well stated earlier).
I want to copy and paste this ten thousand times.
When you don't even 'try' this - you are not engaged with the vehicle.
rush6410
Member
I completely agree. I find it much easier to hold a line in which you expect the car to drive. Then, resist the incorrect AP steer resulting in a disengage maintaining the line. I find it much easier to operate in this manner rather than trying to overpower the correction. I have called this AP defensive driving.
Advanced. Lol. This is how people drive regular cars all the time. This is not rocket science.
I wish Tesla would notify us when a software update touches the AP codebase. In the release notes they could just say "Autopilot software updated, please use caution" or something. That way everyone knows when AP may start behaving differently, even folks who don't read the forums or Electrek or whatever. The fact that the code change on 10.4/10.5 was so huge and it was not mentioned at all kind of boggles my mind.
That's got to be one BIG pedestrian!
The AP 2.0 Radar can’t discriminate targets. The car is not supposed to hit other cars, so it works.
I believe the 2.5 radar can tell the difference between people & vehicles & tell the vehicle. Then it’s up to Tesla to display people instead of cars.
Then why does AP1 know about pedestrians? ME visual recognition and same radar...
You need stereo vision on the front to distinguish objects from the road on the front-facing camera because everything is coming towards you, and so you need to determine if something is actually sticking up from the road meaningfully. However, that's only important for non-moving objects (which would otherwise be indistinguishable from the road except by shape. You ostensibly might want stereo cameras to check the lane coming towards you for non-moving objects, but realistically, if you detect something non-moving after you start into the lane, you can always go back into your lane, so that's mostly a non-issue (though perhaps not entirely).
For the back cameras, you definitely don't care about non-moving objects because you've already passed them. The only thing you care about are objects that are moving faster than you are. You can easily distinguish those from everything else because they're the only objects getting bigger from one frame to the next instead of staying the same or getting smaller. So you shouldn't need stereo cameras in the back at all.
Unless there's a bug that prevents disengaging autopilot. We can't definitively exclude that possibility until they have analyzed the data from the car (and maybe not even then).
Last edited:
One thing that I've noticed over time with AP2 is that the amount of tension in the wheel has varied (how tightly AP "holds" on when trying to disengage). Its been, for the past 5 months, very easy to torque the wheel and disengage (or, as a poster above noted, just hold the wheel when AS turns to resume steering control).
I just want to add one thing to anyone who isn't super experienced with AP (or maybe hasn't had a crisis or occasion to think about disengagements).
On my first real scary AP2 incident when they first released highway AS, AS lost the lane lines and took me straight at the concrete barrier. I overcorrected on turning the wheel because I was inexperienced and almost snapped the car back into a neighboring lane. In recreating the incident in my brain, I realized that when I snapped the wheel, AS disengaged but TACC was still active and so I was still going 70mph.
Since then I've always tapped the brake (easiest and most certain way) and I use a death grip to let AS break itself only when I want to keep the car in control of speed. I've learned my brain doesn't allow me in a crisis to push the stalk forward to disengage (ironically I do this all the time in non-crisis disengagement because its the most seamless and I can feather the accelerator so the passenger wouldn't know except for the tell tale "ding-dong" noise which only rarely interrupts the music). So brake is my go to move (but only a tap unless I hear the FCW).
Anyways, my PSA for all AP users (TLDR) is to think that crisis situation through and just go with whichever you trust. Brake will kill everything (TACC and AS). Pushing the stalk forward does too. Turning the wheel leaves your car in control of the speed (either maintain, accelerate to previously set max, or slow/stop), so use that only when you want the car to maintain that task.
I just want to add one thing to anyone who isn't super experienced with AP (or maybe hasn't had a crisis or occasion to think about disengagements).
On my first real scary AP2 incident when they first released highway AS, AS lost the lane lines and took me straight at the concrete barrier. I overcorrected on turning the wheel because I was inexperienced and almost snapped the car back into a neighboring lane. In recreating the incident in my brain, I realized that when I snapped the wheel, AS disengaged but TACC was still active and so I was still going 70mph.
Since then I've always tapped the brake (easiest and most certain way) and I use a death grip to let AS break itself only when I want to keep the car in control of speed. I've learned my brain doesn't allow me in a crisis to push the stalk forward to disengage (ironically I do this all the time in non-crisis disengagement because its the most seamless and I can feather the accelerator so the passenger wouldn't know except for the tell tale "ding-dong" noise which only rarely interrupts the music). So brake is my go to move (but only a tap unless I hear the FCW).
Anyways, my PSA for all AP users (TLDR) is to think that crisis situation through and just go with whichever you trust. Brake will kill everything (TACC and AS). Pushing the stalk forward does too. Turning the wheel leaves your car in control of the speed (either maintain, accelerate to previously set max, or slow/stop), so use that only when you want the car to maintain that task.
Pro tips:
1) pushing the stalk (flicking forward with style) is the gentleman’s way to gracefully disengage AP...
2) dial in follow length to change AP behavior and hesitancy, follow distance of 6 or 7 causes very cautious adjustment of speed when cars are in adjacent lane.... it’s pretty nifty, 7 is like a nervous adjacent lane setting on 10.4
3) don’t like how long it takes to speed up when on autopilot after a red light or a lead car turns into a parking lot while you continue straight? No worries, don’t disengage, step on the accelerator while on autopilot all the way until you are back up to speed... just don’t keep pressing the pedal after you reach the speed setting
4) smile
Tam
Well-Known Member
With recent 2 blogs for this accident, Telsa's claimed that Autopilot can "reduce crash rates by as much as 40%" as reported by Department of Transportation, Quality Control Systems has sued to get the supporting data and method for such claim.
Last edited:
bonnie
I play a nice person on twitter.
There are countless examples in the medical device & pharma industries of competitive info being withheld from an FOI request, but the manufacturer is still able to make performance & safety claims after their data has been reviewed by a government agency.
Similar threads
