The point of moving to the full stack is heading to at least L3 self driving. As of now the AP stack is probably somewhat legacy and not receiving too much attention. So once it is replaced with single stack on Beta and tested for a few months it will then likely be replaced on all standard AP cars and will vastly improve all safety aspects. So probably be late this year at the earliest before standard AP users get the new stack.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
AP/FSD related crashes
- Thread starter Terminator857
- Start date
Whether or not AP is functional shifts zero liability in either direction, the driver is required to be engaged at all times and will need to be engaged at all times whenever a full stack is released. I'd be surprised if anyone believes Tesla will be taking on liability for this stuff any time in the foreseeable future.
The only implication here will be in what the NHTSA thinks might be required in terms of driver engagement monitoring, in other words increased sensitivity of the monitoring systems or shortening lags between detecting driver disengagement and when warnings are thrown or the system disengages.
Tam
Well-Known Member
From the picture, it doesn't look like a Model S or X so it could not be AP1. It's a newer version for either Model 3 or Y:
Photo: Gephardt Daily/Monico Garza/SLCScanner
Tam
Well-Known Member
It's the 2020 Tesla Model 3. The driver said it's Autopilot, but we won't know for sure until NHTSA gets the car's log.
U.S. probes Tesla crash that killed motorcyclist in Utah
The National Highway Traffic Safety Administration (NHTSA) said on Tuesday it has opened a special investigation into the crash of a 2020 Tesla Model 3 vehicle that killed a motorcyclist in Utah.
www.reuters.com
AlanSubie4Life
Efficiency Obsessed Member
A little hard to tell for sure, but the speed difference looks like it was substantial (more than 15mph maybe?). And it was a somewhat offset collision.
But it did say the motorcyclist was traveling in the lanes, so not stationary.
On the radar question:
We don’t know (if AP was in use) whether this was TeslaVision or not for this vintage. Guess we’ll see if he was traveling more than ~87mph (85mph limit, which locks out at ~87mph)? I think that is correct anyway? If 87mph to 90mph then it had radar.
I guess another way to summarize is only TeslaVision if using FSD Beta?
Most likely TeslaVision gets a pass on this one…I think.
But it did say the motorcyclist was traveling in the lanes, so not stationary.
On the radar question:
We don’t know (if AP was in use) whether this was TeslaVision or not for this vintage. Guess we’ll see if he was traveling more than ~87mph (85mph limit, which locks out at ~87mph)? I think that is correct anyway? If 87mph to 90mph then it had radar.
I guess another way to summarize is only TeslaVision if using FSD Beta?
Most likely TeslaVision gets a pass on this one…I think.
I don't believe a NHTSA ADAS report would be conclusive. As I recall, there is no submission field for the timing between AP disconnect and the crash. All you can say was that AP was in use within 30 secs of the crash. That's a long time.It's the 2020 Tesla Model 3. The driver said it's Autopilot, but we won't know for sure until NHTSA gets the car's log.
U.S. probes Tesla crash that killed motorcyclist in Utah
The National Highway Traffic Safety Administration (NHTSA) said on Tuesday it has opened a special investigation into the crash of a 2020 Tesla Model 3 vehicle that killed a motorcyclist in Utah.
Of course, if the driver was lying, and AP was not in use at all, then there will be no report.
But then, NHTSA might get better data for a special investigation...
Tam
Well-Known Member
2020 Tesla Model 3 does have radar, but the driver needs FSD beta to disable the radar.
Tam
Well-Known Member
This article cited that the 2020 Tesla Model 3 merged into the HOV lane then hit the Harley Davidson.
I am not sure that do they mean "changed lane", and not "merged".
I think merging task is more difficult than Auto Lane Change. I don’t think Tesla System is that good in the merging maneuver and collisions can happen.
If the driver doesn't have EAP/FSD/FSD beta then there’s no Auto Lane Change. In that case, it's a manual lane change and the driver didn't notice the Harley Davidson.
That motorcycle can make loud noise so it might take lots of distraction to not hearing it.
www.dailymail.co.uk
I am not sure that do they mean "changed lane", and not "merged".
I think merging task is more difficult than Auto Lane Change. I don’t think Tesla System is that good in the merging maneuver and collisions can happen.
If the driver doesn't have EAP/FSD/FSD beta then there’s no Auto Lane Change. In that case, it's a manual lane change and the driver didn't notice the Harley Davidson.
That motorcycle can make loud noise so it might take lots of distraction to not hearing it.
Biker killed after Tesla on Autopilot smashes into motorcycle
A motorcyclist was killed by a Tesla driving on Autopilot on a Utah highway early Sunday morning, the latest in a string of recent fatal accidents involving self-driving technology.
AlanSubie4Life
Efficiency Obsessed Member
Even if this was not on AP, I really think Tesla should roll the new driver monitoring to all Tesla AP users with hardware that supports it and tell them to suck it up. It seems quite a bit better at making sure people are paying attention, even if it is not perfect.
Need to be proactive. Plus people and customers would complain a lot (privacy, etc.), which is a good sign that it is the right thing to do.
Don’t have to completely disable AP, but could have fewer strikes and just disable until park as usual. If they do this for excessive speed (they do over 90mph as I recall) it’s an even better idea for not paying attention.
I’m surprised they have not already, given their focus on safety. It’s a top priority!
I looked up the location on Streetview (lost the link and the exact sign pictured wasn’t 100% clear), but in all cases it was a completely trivial situation. Straight, well marked, with simple single solid white line marking the carpool lane. Basically impossible to have an accident (even if the lights were out on the bike - there appeared to be large light poles as well).
Need to be proactive. Plus people and customers would complain a lot (privacy, etc.), which is a good sign that it is the right thing to do.
Don’t have to completely disable AP, but could have fewer strikes and just disable until park as usual. If they do this for excessive speed (they do over 90mph as I recall) it’s an even better idea for not paying attention.
I’m surprised they have not already, given their focus on safety. It’s a top priority!
I looked up the location on Streetview (lost the link and the exact sign pictured wasn’t 100% clear), but in all cases it was a completely trivial situation. Straight, well marked, with simple single solid white line marking the carpool lane. Basically impossible to have an accident (even if the lights were out on the bike - there appeared to be large light poles as well).
Last edited:
Yeah I agree but their options were limited, waving the flashlight ultimately did not work for all the cars. Standing on the road or pointing directly at drivers or at the flipped car may not have worked for everyone either.
I once tried to wave down people from driving at full speed through a busy six lane four-way traffic light that had lost power. Amazingly most drivers were going through at full speed in all directions. I tried to make people aware from my direction but it made no difference. I called 911, asked for a traffic cop and walked away. I had no other tools and was unwilling to further risk my life when it clearly wasn't working. I didn't stay and watch because if a crash happened I'd feel bad, but the situation was beyond my control.
Last edited:
Not to mention the fact that they are the insurance agencies for quite a few of them.
Why would you not?
I think the real reason is it driver monitoring simply doesn't work well enough to roll it out to the masses.
AlanSubie4Life
Efficiency Obsessed Member
I’m not sure about that at all. It seems to work perfectly fine for FSD Beta. Never any issues (N=1). Obviously they would have to fix all the broken cabin cameras but that would be quick since they can usually detect it.
I think it is a big improvement over the existing setup. I guess compared to that - it sounds like you are saying it is worse?
I can think of other reasons why they would not want to push it (there would be a LOT of complaints when requiring camera activation to use a feature which did not previously require that). To me that seems more likely?
Last edited:
and to take your logic a step further: automakers don't prioritize safety and accident prevention because insurance pays for everything.He's said this multiple times in interviews and tweets. Yes AP/FSD is imperfect today and can/will crash without proper supervision, but to conclude that Tesla doesn't prioritize collisions is ridiculous.
Tam
Well-Known Member
Prioritize in talk but not in action. It made things worse by taking the radar away and triggering a firmware recall due to false Forward Collision Warnings and Automatic Emergency Brakes.
He's said this multiple times in interviews and tweets. Yes AP/FSD is imperfect today and can/will crash without proper supervision, but to conclude that Tesla doesn't prioritize collisions is ridiculous.
Remember, according to the talk, the Tesla system is not immature. It has matured enough during the Earnings Call last week as it tried to guilt those not paying $12,000 because the price will increase as "It does seem as though we are converging on solving full self-driving this year."
Last edited:
I was going off the complaints posted on TMC along with my own N=1 experience.
In my own testing it didn't seem to work all that well. Like I could be playing with the screen and it never nagged me or told me to pay attention. I don't think it ever issued a false positive, but was blissfully ignorant (false negative) when I fought with navigation.
As end users we have no idea how well it does across the user base, and across different lighting situations.
It would be a pretty amazing accomplishment if the cabin camera did work well for driver monitoring given the fact that it was never intended for driver monitoring. Now there is an argument to be made that it doesn't necessarily have to work that well as long as false positives are kept really low while also triggering on obvious things like texting.
All we really know is its used in FSD Beta, but only partially. It isn't the sole driver engagement feedback as the torque sensor is still used despite not being the best method for city streets with lots of turns.
Given the level of scrutiny Tesla is under I think the userbase would understand requiring the camera given the fact that there really wasn't that much upheaval when they activated it awhile ago.
In fact if regular owners were asked they might assume its already being used or they had no idea it was even there.
In my Rivian the driver monitoring camera isn't activated yet. They just use the capacitive sensors.
AlanSubie4Life
Efficiency Obsessed Member
Try looking out the side window for a bit (being sure to keep things safe by keeping eyes forward as much as possible, and have a spotter if you want to be even more safe). It will trigger.
I definitely don’t think it is perfect, but my impression is that it is a distinct improvement. As you say, as long as it provides more true positives and does not increase false positives, it is better. False negatives are what they are…. There is no question that you can get away with texting with the new system if you are determined. I am not claiming it solves all problems; I don’t think any monitoring system is able to do that.
KArnold
Active Member
Makes me wonder if the same car/situation had happened but NOT on AP - would there have been anything different? We'll never know but I'd guess likely not especially with the distraction of flashlight guy.
That's not a reason to not pursue reasonable precautions on AP of course. But even with a well designed AEB (as most new cars already have), there will be a accidents like this. Appears to be difficult to prevent entirely but AEB certainly should have been on the binders - that's the failure here.
That's not a reason to not pursue reasonable precautions on AP of course. But even with a well designed AEB (as most new cars already have), there will be a accidents like this. Appears to be difficult to prevent entirely but AEB certainly should have been on the binders - that's the failure here.
AlanSubie4Life
Efficiency Obsessed Member
I absolutely think AP was a factor here. Easy to lose situational awareness and attentiveness. Use of AP also nearly invariably delays the onset of braking or easing off on the accelerator (occurs naturally with a human when encountering uncertainty) which would have made a big difference here. The overturned vehicle was likely much more visible in reality than it appears in the video from the FSD cameras.
We’ll never know though.
Similar threads
