How Tesla can defeat AP-Buddy-style devices on AP, EAP and FSD at L2

[OPRCE]

Someone posted a [since deleted] thread this morning entitled “Handsfree Autopilot for $5”, containing just a photo of a water-bottle lashed to the steering, possibly in a misplaced effort at humour but it got me thinking how to make the system more foolproof before anyone [else] goes killing themselves or, worse, someone else, by abusing the increasing capabilities of AP.

Idea: require the torque-check on wheel before each lane-change in AP to operate in the same sense as the indicator.

Currently torque in either direction will suffice for a lane-change to proceed, whether initiated manually via indicator or by Unconfirmed Lane-Change under Nav-on-AP since v.2019.8.5, but this simple software fix can immediately eliminate 99% of the attraction in using defeat devices (weight on the steering wheel, as illustrated), as those will always produce torque in the same direction, whereas lane-changes will average out at 50:50 to each side.

Hence, whether in basic AP mode, in EAP using ULC in NoAP or soon the feature-complete supervised-FSD, this intent-aware torque-sensing would
1. be a more reliable check that active hands are in fact on the wheel
2. additionally prove to the system that the supervising driver is attentive enough to know to which side the car will next shift
3. naturally encourage continued use of the mirrors before AP manoeuvres
4. save being pulled over/prosecuted for careless driving once cops cotton on to the defeat-device trick

I personally think that if only one person is spared GBH by this measure between now and FSD gaining L3 approval, it will have hastened that day and be well worth it, or in any case should help burnish Tesla's reputation as a company striving to enhance safety.

Anyhow, am interested to hear what you all think ...

 

[electronblue]

The problem with this idea — assuming it technically works — is that it would also defeat the main method of nagless driving with a single hand on the lower portion of the wheel. While less safe than driving with two hands that gets nags all the time at least there is a hand on the wheel... this would nag even that, right?

 

[OPRCE]

The problem with this idea — assuming it technically works — is that it would also defeat the main method of nagless driving with a single hand on the lower portion of the wheel. While less safe than driving with two hands that gets nags all the time at least there is a hand on the wheel... this would nag even that, right?

No, one hand on the wheel can still briefly torque to whichever side is being indicated, and the same applies to two hands.

This check is only performed when blinker starts (or just before a lane-change), thus if driver is attentive it should produce no added delay in execution and would not affect usual habits to achieve nagless driving at any other time.

PS: Good point though on the technical possibility that the torque sensor cannot differentiate between the directions, tbh I presumed it would be something like a resolver which could read rotation-pressure both ways. Maybe someone knows how it actually works?

 

[electronblue]

No, one hand on the wheel can still briefly torque to whichever side is being indicated, and the same applies to two hands.

This check is only performed when blinker starts (or just before a lane-change), thus if driver is attentive it should produce no added delay in execution and would not affect usual habits to achieve nagless driving at any other time.

Sure. I was mainly commenting on this idea from the perspective of defeating defeat devices without adding to nags — which I feel would be important for its acceptance.

The one hand drive — as bad as it is — is one reason the nags get accepted because it at least often allows escaping the nags.

 

[ewoodrick]

First, the fact that you felt the need to demonstrate yet another nag override demonstration really undermines what your intent may be. Just a simple post would have been one thing, but adding a "how to override nag" is another.

Tesla can easily make the nag override harder, that's easy. But that's not there intent. They don't want nag, they don't want lawsuits either. We've got videos of people falling asleep at the wheel, people driving drunk, people taking videos with no one at the wheel. These things do nothing but undermine what Tesla is trying to do.

And how would you expect "intent aware" solutions to function on 20 mile long straight and flat highways?

 

[OPRCE]

First, the fact that you felt the need to demonstrate yet another nag override demonstration really undermines what your intent may be. Just a simple post would have been one thing, but adding a "how to override nag" is another.

Tesla can easily make the nag override harder, that's easy. But that's not there intent. They don't want nag, they don't want lawsuits either. We've got videos of people falling asleep at the wheel, people driving drunk, people taking videos with no one at the wheel. These things do nothing but undermine what Tesla is trying to do.

And how would you expect "intent aware" solutions to function on 20 mile long straight and flat highways?

1. Sorry if the photo creates an unfortunate impression; it was intended merely to illustrate the kind of recklessness I wish to render a thing of the past, absolutely not to encourage it. On further thought though I agree it is unnecessary, so to eliminate any confusion have requested Mods to assist by deleting the photo and the text ", as illustrated".

2. My proposal surely assists in two important aims for Tesla, i.e. to tangibly increase safety for all users, without adding to the sum of the nags, and second to hopefully shorten the Beta period until they can be dispensed with entirely.

3. On long straight flats without overtaking the system would still function exactly as it currently does. "Intent-aware" does not mean it needs to read the drivers mind or the road conditions, just that the torque required to initiate a lane-change in AP must be congruent with the intention already expressed on the indicator by human or machine.

 

[Kilotango74]

Why doesn't Tesla utilize that camera that already inside the Model 3 to ensure the driver is attentive while AP is engaged. That would cut down on the oranges in the steering wheels and idiots sleeping in the back seat.
"This is where Subaru’s DriverFocus comes in. It monitors the driver for signs of distraction or fatigue. If the system determines that you’re driving while sleepy or distracted, it issues a warning, and can even stop the vehicle if this warning is ignored" The type of implementation could just as easily disengage autopilot if it senses the driver not being attentive.

 

[electronblue]

Why doesn't Tesla utilize that camera that already inside the Model 3 to ensure the driver is attentive while AP is engaged. That would cut down on the oranges in the steering wheels and idiots sleeping in the back seat.
"This is where Subaru’s DriverFocus comes in. It monitors the driver for signs of distraction or fatigue. If the system determines that you’re driving while sleepy or distracted, it issues a warning, and can even stop the vehicle if this warning is ignored" The type of implementation could just as easily disengage autopilot if it senses the driver not being attentive.

One problem could be the camera is not very optimally positioned to ensure driver attention and it is possible that was never really its intended use but instead as said to monitor Tesla Network usage (and crash liability photos).

Of course they could still try to use it for real-time driver attention monitoring, maybe it would work, but this is my current thinking on why they haven’t.

Maybe we’ll learn more on the 22nd.

 

[Kilotango74]

One problem could be the camera is not very optimally positioned to ensure driver attention and it is possible that was never really its intended use but instead as said to monitor Tesla Network usage (and crash liability photos).

Of course they could still try to use it for real-time driver attention monitoring, maybe it would work, but this is my current thinking.

That true, but I think a re-purpose for this type of control would be a much better use of technology than the hey lets turn my car into a ride share auto driving taxi thingy application.

 

[electronblue]

That true, but I think a re-purpose for this type of control would be a much better use of technology than the hey lets turn my car into a ride share auto driving taxi thingy application.

I’m sure it would be appreciated as an alternative to the torque sensing yes.

 

[Kilotango74]

I’m sure it would be appreciated as an alternative to the torque sensing yes.

It just seems to me that the system that Tesla is utilizing currently is just going to lead to legislation that severely curtails its usability. One person screws up we all pay for it.....

 

[OPRCE]

Why doesn't Tesla utilize that camera that already inside the Model 3 to ensure the driver is attentive while AP is engaged. That would cut down on the oranges in the steering wheels and idiots sleeping in the back seat.
"This is where Subaru’s DriverFocus comes in. It monitors the driver for signs of distraction or fatigue. If the system determines that you’re driving while sleepy or distracted, it issues a warning, and can even stop the vehicle if this warning is ignored" The type of implementation could just as easily disengage autopilot if it senses the driver not being attentive.

1. That won't work in Models S/X without an expensive recall/retrofit, which is a whole 'nother ball of wax, whereas my way should be trivial in software and work for all AP vehicles.

2. Also the M3 camera was not designed for this DAM function, in terms of its internals and placement. The best solution so far (cf Cadillac CT6) is a dedicated IR illuminator and corresponding camera pointing directly face-on from below eye-horizon line, which can track eye focus through sun-glasses, under hats, etc., whereas the M3 internal fisheye meant for surveilling eventual RoboTaxi passengers is too high and off-axis so would presumably have the driver's face on the edge of FOV where most lens distortion occurs, and almost certainly captures in the visual spectrum only.

3. It probably could usefully detect someone nodding off in daylight, but may not work at night or for someone who likes to sit up straight but close their eyes for 10 minute stretches behind shades.

4. From the tweets thus far, Musk seems inclined only to allow it to function as an internal SentryCam, which it can do well.

It just seems to me that the system that Tesla is utilizing currently is just going to lead to legislation that severely curtails its usability. One person screws up we all pay for it.....

5. My suggestion to tighten things up could possibly help avert a more restrictive DAM legislation which I agree is more likely [at least in Europe] to happen the longer the system exists as L2 ADAS.

 

[brkaus]

Of course, the iPad on the dash isn’t detected by the Cadillac version.

There is always a cheat.

 

[OPRCE]

It just seems to me that the system that Tesla is utilizing currently is just going to lead to legislation that severely curtails its usability. One person screws up we all pay for it.....

Of course, the iPad on the dash isn’t detected by the Cadillac version.

There is always a cheat.

Yes, as they say, every foolproof system will eventually produce a superior fool, but Cops would pretty soon haul over anyone they spot with an iPad atop the dash streaming video, as it illegally obstructs the view forward and severely distracts from the driver's duty of care, so that is slightly outside the scope here.

 

[bikeandsail]

You should call this the suicide device

 

[Nocturnal]

Why not just look at the variance of torque? A defeat device will have almost no variability, an arm, hand, leg will have a ton of variability.

 

[OPRCE]

Why not just look at the variance of torque? A defeat device will have almost no variability, an arm, hand, leg will have a ton of variability.

Good point, with a few more cycles of CPU this should also work.

I would then say do both.

 

[Nocturnal]

Good point, with a few more cycles of CPU this should also work.

I would then say do both.

If the Xbox connect I had 10 years ago could track my head reasonably well then surely the camera above the mirror can. (just like Cadillac does)

 

[OPRCE]

If the Xbox connect I had 10 years ago could track my head reasonably well then surely the camera above the mirror can. (just like Cadillac does)

See comment #12 above.

 

[iTech]

If the Xbox connect I had 10 years ago could track my head reasonably well then surely the camera above the mirror can. (just like Cadillac does)

Not all of the models have that.