Oils4AsphaultOnly
Posts: 669
Joined: Sat Oct 10, 2015 4:09 pm
Delivery Date: 20 Nov 2016
Leaf Number: 313890
Location: Arcadia, CA

Re: Tesla's autopilot, on the road

Thu Sep 05, 2019 6:14 pm

GRA wrote:
Thu Sep 05, 2019 5:26 pm
The driver had his hands off the wheel for all but 51 seconds of the last 14 minutes A/P was engaged, and either had a bagel or a cup of coffee in one hand, plus was distracted by the radio. The report goes into detail about how the A/P warning system works, including the fact that on a divided highway you can go 3 minutes hands-off before a warning is issued, if there is a lead car, and 2 minutes otherwise. The first audible warning is issued 15 seconds after the visual warning, the second 10 seconds after that, and the third 5 seconds later, at which point the car begins to decelerate. In other words, you can go at least (see below) 3 min. 30 seconds with your hands off the wheel before the car will assume no one's paying attention.

The car had hardware v1.0 and firmware 17.50.97-3bd9f6d082U, installed on 12/28/2017. Some of the warning times have apparently changed in later versions of hardware and firmware. The total trip was 66 minutes, and A/P was engaged for a total of 29 min. 4 sec., with hands detected on the wheel for only 78 sec. total. During the last A/P segment, visual hands-off alerts were issued four times,, and no hands were detected for the last 3'41" before impact. This was possible because the car's speed dropped below 25 mph at some point, which allows up to 10 minutes before a warning is issued. The Tesla slowed to 21 mph behind the lead vehicle.

When the lead vehicle changed lanes to avoid the fire truck 3-4 seconds before impact, the Tesla began to accelerate to the TACC set speed of 80 mph (speed limit is 65). At 0.49 sec. before impact the FCW detected a stationary object (the fire engine) and issued audible and visual warnings to the driver. The Tesla was doing 30.9 mph at impact. AEB didn't activate nor would it as it's known that it can't deal with this situation.

To what should be no one's surprise, this is another confirmation that A/P as implemented not only allows but even encourages driver disengagement, just as all the research indicated would be the effect of such an implementation.
Lovely how you're still on your high-horse without any expertise in AI whatsoever.

Isn't it about time for another A/P death? Yet, there hasn't been one since March? Why focus so intently on the accidents that have happened, and NOT on the ones avoided? Those news reports about drivers caught sleeping on A/P were all instances of drivers that could've died if they were in another car. A/P isn't perfect. Focusing on the accidents that have occured is completely missing the point of A/P.
:: Model 3 LR :: acquired 9 May '18
:: Leaf S30 :: build date: Sep '16 :: purchased: Nov '16
100% Zero transportation emissions (except when I walk) and loving it!

GRA
Posts: 10724
Joined: Mon Sep 19, 2011 1:49 pm
Location: East side of San Francisco Bay

Re: Tesla's autopilot, on the road

Sun Sep 08, 2019 3:39 pm

Oils4AsphaultOnly wrote:
Thu Sep 05, 2019 6:14 pm
GRA wrote:
Thu Sep 05, 2019 5:26 pm
The driver had his hands off the wheel for all but 51 seconds of the last 14 minutes A/P was engaged, and either had a bagel or a cup of coffee in one hand, plus was distracted by the radio. The report goes into detail about how the A/P warning system works, including the fact that on a divided highway you can go 3 minutes hands-off before a warning is issued, if there is a lead car, and 2 minutes otherwise. The first audible warning is issued 15 seconds after the visual warning, the second 10 seconds after that, and the third 5 seconds later, at which point the car begins to decelerate. In other words, you can go at least (see below) 3 min. 30 seconds with your hands off the wheel before the car will assume no one's paying attention.

The car had hardware v1.0 and firmware 17.50.97-3bd9f6d082U, installed on 12/28/2017. Some of the warning times have apparently changed in later versions of hardware and firmware. The total trip was 66 minutes, and A/P was engaged for a total of 29 min. 4 sec., with hands detected on the wheel for only 78 sec. total. During the last A/P segment, visual hands-off alerts were issued four times,, and no hands were detected for the last 3'41" before impact. This was possible because the car's speed dropped below 25 mph at some point, which allows up to 10 minutes before a warning is issued. The Tesla slowed to 21 mph behind the lead vehicle.

When the lead vehicle changed lanes to avoid the fire truck 3-4 seconds before impact, the Tesla began to accelerate to the TACC set speed of 80 mph (speed limit is 65). At 0.49 sec. before impact the FCW detected a stationary object (the fire engine) and issued audible and visual warnings to the driver. The Tesla was doing 30.9 mph at impact. AEB didn't activate nor would it as it's known that it can't deal with this situation.

To what should be no one's surprise, this is another confirmation that A/P as implemented not only allows but even encourages driver disengagement, just as all the research indicated would be the effect of such an implementation.
Lovely how you're still on your high-horse without any expertise in AI whatsoever.

Isn't it about time for another A/P death? Yet, there hasn't been one since March? Why focus so intently on the accidents that have happened, and NOT on the ones avoided? Those news reports about drivers caught sleeping on A/P were all instances of drivers that could've died if they were in another car. A/P isn't perfect. Focusing on the accidents that have occured is completely missing the point of A/P.

I'd written a long reply to this, but MNL timed me out and made me log-in again, and the reply was gone after I did so. As I lack the patience or interest to re-write it all, this will have to do, which covers some but not all of the points I did:
Feds scold Tesla for slow response on driver monitoring
In 2017, NTSB called steering wheel torque a "poor surrogate" for driver attention.
https://arstechnica.com/cars/2019/09/fe ... onitoring/


Eye-monitoring cameras have proven to be more effective than torque sensors at ensuring drivers remain engaged, and ideally you want to use both systems - Cadillac does so, and so do several other manufacturers.
Guy [I have lots of experience designing/selling off-grid AE systems, some using EVs but don't own one. Local trips are by foot, bike and/or rapid transit].

The 'best' is the enemy of 'good enough'. Copper shot, not Silver bullets.

GRA
Posts: 10724
Joined: Mon Sep 19, 2011 1:49 pm
Location: East side of San Francisco Bay

Re: Tesla's autopilot, on the road

Tue Sep 10, 2019 12:28 pm

Assuming this isn't fake, I think this is a first, with both the driver and passenger asleep - Arstechnica:
Another Tesla driver apparently fell asleep—here’s what Tesla could do
Tesla needs a better driver monitoring system.
https://arstechnica.com/tech-policy/201 ... r-problem/


Video here: https://www.cbsnews.com/news/tesla-driv ... n-highway/
Guy [I have lots of experience designing/selling off-grid AE systems, some using EVs but don't own one. Local trips are by foot, bike and/or rapid transit].

The 'best' is the enemy of 'good enough'. Copper shot, not Silver bullets.

Oils4AsphaultOnly
Posts: 669
Joined: Sat Oct 10, 2015 4:09 pm
Delivery Date: 20 Nov 2016
Leaf Number: 313890
Location: Arcadia, CA

Re: Tesla's autopilot, on the road

Tue Sep 10, 2019 3:40 pm

GRA wrote:
Tue Sep 10, 2019 12:28 pm
Assuming this isn't fake, I think this is a first, with both the driver and passenger asleep - Arstechnica:
Another Tesla driver apparently fell asleep—here’s what Tesla could do
Tesla needs a better driver monitoring system.
https://arstechnica.com/tech-policy/201 ... r-problem/


Video here: https://www.cbsnews.com/news/tesla-driv ... n-highway/
These are people who would've been driving tired and exhausted. Isn't it nice that they didn't hit anyone?
:: Model 3 LR :: acquired 9 May '18
:: Leaf S30 :: build date: Sep '16 :: purchased: Nov '16
100% Zero transportation emissions (except when I walk) and loving it!

GRA
Posts: 10724
Joined: Mon Sep 19, 2011 1:49 pm
Location: East side of San Francisco Bay

Re: Tesla's autopilot, on the road

Wed Sep 11, 2019 5:52 pm

Oils4AsphaultOnly wrote:
Tue Sep 10, 2019 3:40 pm
GRA wrote:
Tue Sep 10, 2019 12:28 pm
Assuming this isn't fake, I think this is a first, with both the driver and passenger asleep - Arstechnica:
Another Tesla driver apparently fell asleep—here’s what Tesla could do
Tesla needs a better driver monitoring system.
https://arstechnica.com/tech-policy/201 ... r-problem/


Video here: https://www.cbsnews.com/news/tesla-driv ... n-highway/
These are people who would've been driving tired and exhausted. Isn't it nice that they didn't hit anyone?

These are people who may have decided to drive tired and exhausted because they had A/P. Is that safer than not driving? We can certainly be glad that they didn't encounter a stopped vehicle in their lane, as they were apparently traveling at 55-60 mph, unlike many of the other Tesla crashes into non-moving vehicles - the fatal crash in China was moving at high speed. More importantly, IIRR so far no one has been in/on any of the vehicles that A/P-guided Teslas have rear-ended; if a firefighter/police officer/maintenance worker had been in or standing on or behind their stopped vehicle at the time and was injured or killed, would you be so blase' about this? Do you think the public will be? I mean, the occupants decided to do something stupid and thus consented to the risk, but other members of the public haven't given any such consent.

Beyond that, does this and numerous other examples of misusing A/P indicate that the system as implemented does the best possible job of ensuring that the driver remains alert and ready to instantly retake control when needed, given the design options and equipment available to Tesla? Of course it doesn't.

Would drastically shortening the time to the first hands-off warning and the following slow to a stop make drivers keep their hands on the wheel and pay more attention to avoid constant annoying warnings, especially if they were both visual and aural? Sure. Google used 6 seconds for their driver-assist program (before deciding it wasn't worth the effort and going for FSD), but for those companies who do want to offer driver assistance, this is an obvious step, far better than the minutes that Tesla currently allows.

Would adding an eye monitoring camera further increase the chances that a driver will be watching the road instead of watching a movie or sleeping, and oblivious to everything that's going on around them? You bet.
Guy [I have lots of experience designing/selling off-grid AE systems, some using EVs but don't own one. Local trips are by foot, bike and/or rapid transit].

The 'best' is the enemy of 'good enough'. Copper shot, not Silver bullets.

Return to “Off-Topic”