Tesla Working on Driver Monitoring System

[Needsdecaf]

Masking tape (or a clever clip design) solves it for all who want to maintain privacy. The trade-off could be their safety by also defeating automation features. But ultimately their choice and associated higher insurance rates that would likely follow. Funny how the system will work itself out there. Indirectly they'd be paying for their own privacy much like paying to stop those annoying ads, both being your choice to make.

Higher insurance rates? Reducing safety by defeating automation features? What on earth are you on about here?

TACC / Autosteer use not tied to any safety systems. Lane departure, obstacle avoidance, AEB all still work whether TACC engaged or not.

Also, insurance rates based on the car's features, not whether you use them or not. If I own a BWM with their Full Active Driving Safety features, my insurance has no idea if I get in and hit the OFF button when I drive. I suppose if I got in an accident and they decided to subpoena BMW and get the black box that detects whether it's on or off they would know that answer, but my Auto Insurance policy doesn't require me to keep any of that switched on so what would be the point?

BTW, GM Cruise has been tracking eyes-on-the-road as well for years, but I've never seen anything in the media to challenge driver privacy there. That's a double standard to all those pressing the "Publish" button in the sold-out media, for more loss of trust.

So does BMW and so do others. But they use the same technology built into your iphone: facial tracking through IR sensors. That's not going to see what your hands are doing, just where your eyes are pointed. They don't use a camera that can see everything else that goes on in your cabin.

I'm glad you decided that it's no big deal to you. That's 100% your choice and I respect that. When Tesla sold the Model 3, it was with a camera that was "not in use" and they never disclosed that it could be used for TACC / Autosteer monitoring. I'd have a problem if they suddenly tied the camera to those functions.

If you recall, Tesla explicitly added the ability to use the interior camera for this purpose, and there is a setting to turn it off if you are uncomfy with it (the release notes went into this in detail when it was added). So I dont really see this as "big brother" (yet). And if you are REALLY worried and/or paranoid, just put some tape over the camera hole.

No, they never explicity stated anything about the use of the camera "for this purpose" (which was Autopilot according to the quote you posted).

 

[Needsdecaf]

I have one of those little sliding covers over my camera. The problems start if they decide to force you to allow the camera to record in order to be able to use the driver assistance features you paid for ...

Regarding privacy, I have very little confidence in Tesla. They still don't offer a way on their web site for California residents to exercise their rights according to the CCPA. In their privacy policy they basically claim the right to collect whatever telemetry data they want. In the past they have also pulled telemetry off cars that got into accidents and published the data without the owner's permission, using it against the owner (e.g. in case of some Autopilot accidents).

100% this.

 

[verygreen]

lol @ mask inside the car - alone.

some people don't want to broadcast their face across the internet, connected to their name at that.

 

[Dolemite]

If this turns the driving experience into a nanny-state-esque pain in the ****ing ass, I'm immediately selling this car & buying a used F80 M3. Real talk.

 

[vickh]

If Tesla did use the cabin camera for driver attention and implement hands-free driver monitoring, I'd be incredibly happy.

same here. And if you want privacy keep a hand on wheel.

 

[sleepydoc]

If Tesla did use the cabin camera for driver attention and implement hands-free driver monitoring, I'd be incredibly happy.

Same here. My iPhone does a very good job of judging where my attention is; if Tesla could do something similar it would be far better than the system they have now.

Tesla can go F off with more nanny software.

The tape aint coming off.

No one is forcing you to use it, but at the risk of offending - based on the irresponsible behavior we've seen and how unreliable drivers are at paying attention, anyone who wants to use autopilot without some sort of monitoring software can go F off.

Seriously, the auto pilot software is still evolving and is far from perfect. The software needs a nanny, but we've seen far to many cases of people completely misusing it (including someone crawling into the passenger seat.) Unfortunately, the 'nanny' software is necessary to protect other drivers and for Tesla to protect itself.

 

[drtimhill]

No one is forcing you to use it, but at the risk of offending - based on the irresponsible behavior we've seen and how unreliable drivers are at paying attention, anyone who wants to use autopilot without some sort of monitoring software can go F off.

Seriously, the auto pilot software is still evolving and is far from perfect. The software needs a nanny, but we've seen far to many cases of people completely misusing it (including someone crawling into the passenger seat.) Unfortunately, the 'nanny' software is necessary to protect other drivers and for Tesla to protect itself.

Although I agree with this, I'm concerned with the assumption that a camera-base system will inherently be better (or even as good as) the current wheel-nudge system. Intuitively, I think many people assume it would be, one reason being that it would be harder to cheat. However, intuition can be a poor guide, and I've not seen any decent scientific comparisons between the two systems. Sure, an eye system (if working well) can tell you are looking forward, but can it tell you are paying attention? You might be staring off into the middle distance while thinking about what to cook for dinner. One advantage of the wheel-nudge is it requires active input from the driver ("are you there?") as opposed to passive deduction ("I think the driver is there"). There is a good reason trains have an active attention system (dead-mans handle) even today.

And you can be sure people will figure out ways to cheat any eye-checking system, dont underestimate the ingenuity of irresponsible idiots.

 

[hellchris]

Although I agree with this, I'm concerned with the assumption that a camera-base system will inherently be better (or even as good as) the current wheel-nudge system. Intuitively, I think many people assume it would be, one reason being that it would be harder to cheat. However, intuition can be a poor guide, and I've not seen any decent scientific comparisons between the two systems. Sure, an eye system (if working well) can tell you are looking forward, but can it tell you are paying attention? You might be staring off into the middle distance while thinking about what to cook for dinner. One advantage of the wheel-nudge is it requires active input from the driver ("are you there?") as opposed to passive deduction ("I think the driver is there"). There is a good reason trains have an active attention system (dead-mans handle) even today.

And you can be sure people will figure out ways to cheat any eye-checking system, dont underestimate the ingenuity of irresponsible idiots.

I used OpenPilot on a Civic for two years and about 25k miles (system engaged) before I got my Model 3. I can tell you anecdotally that I feel like my Model 3 lets me get away with dangerous driving for way too long. All you have to do is hang your dead weight on the wheel and you can divert your attention for as long as you want. I'm guilty of this when messing with the questionable media controls.

OP only gives you a few seconds of diverted gaze before sounding alerts. Even if I was looking forward and thinking about something else, my lizard brain would still instantly respond to threats in my field of vision. I feel like this is the important part. You can't respond to what you don't see coming and a system that makes you keep your eyes ahead is far better than one that requires a little weight on the wheel every half mile. This is just my experience, but I felt much more engaged with OP than I do with AP.

 

[EVNow]

OP only gives you a few seconds of diverted gaze before sounding alerts. Even if I was looking forward and thinking about something else, my lizard brain would still instantly respond to threats in my field of vision.

Since lizards don’t drive, it won’t. I once hit a curb when parking as I was distracted by talking on phone (hands free).

 

[hellchris]

Since lizards don’t drive, it won’t. I once hit a curb when parking as I was distracted by talking on phone (hands free).

I'm sorry, I thought I could use common expressions to communicate with you. Please understand, I am not speaking in hypotheticals. I have actual experience with both systems. I also don't know what your manual distracted driving has to do with the self driving systems we are talking about. Poor control of your vehicle can only be fixed with practice, not a driver monitoring system.

We have "hands off, eyes on" (OP) and "hands on, eyes wherever you want" (AP). Eyes on the road is safer no matter where you have to stick your hands. You shouldn't need a peer-reviewed study to accept this. You can't respond to things you don't observe, no matter which part of your cortex you are engaging.

 

[drtimhill]

I'm sorry, I thought I could use common expressions to communicate with you. Please understand, I am not speaking in hypotheticals. I have actual experience with both systems. I also don't know what your manual distracted driving has to do with the self driving systems we are talking about. Poor control of your vehicle can only be fixed with practice, not a driver monitoring system.

We have "hands off, eyes on" (OP) and "hands on, eyes wherever you want" (AP). Eyes on the road is safer no matter where you have to stick your hands. You shouldn't need a peer-reviewed study to accept this. You can't respond to things you don't observe, no matter which part of your cortex you are engaging.

The point is you do need a peer reviewed study. You may feel intuitively that your experience was better with OP, and it may have been, for you. Can you extrapolate from this one data point to every driver, in all conditions? Are your intuitive feelings correct? I’m not saying one system is better than the other, I’m saying we don’t know until some one does more rigorous, scientific, research.

And let’s not lose sight that, at the end of the day, drivers are deemed responsible for safety. There are subtle liability issues when you install “driver attention” systems. Sure, to an extent, the manufacturer is protecting themselves from liability, but they are also exposing themselves. I’m waiting for the first lawsuit from a driver claiming that “the car was supposed to ensure I was paying attention, so it was the cars fault”.

 

[EVNow]

I'm sorry, I thought I could use common expressions to communicate with you. Please understand, I am not speaking in hypotheticals. I have actual experience with both systems. I also don't know what your manual distracted driving has to do with the self driving systems we are talking about. Poor control of your vehicle can only be fixed with practice, not a driver monitoring system.

We have "hands off, eyes on" (OP) and "hands on, eyes wherever you want" (AP). Eyes on the road is safer no matter where you have to stick your hands. You shouldn't need a peer-reviewed study to accept this. You can't respond to things you don't observe, no matter which part of your cortex you are engaging.

Dude, I hit the curb once in 20 years - “poor control” is not the reason.

I completely understand what you are trying to say with lizard brain. I’m just saying that’s incorrect - you need to be actively engaged to take quick evasive/corrective car actions. The old regions of brain only help you to duck if a projectile is coming towards you, not to steer the car or brake quickly.

 

[hellchris]

I’m saying we don’t know until some one does more rigorous, scientific, research.

Not we, you. I do know. I'm wondering why you think it would at all be worse to make the driver look forward vs not. You really think AP is safe if you keep your hand on the wheel but close your eyes? (of course you don't)

I completely understand what you are trying to say with lizard brain. I’m just saying that’s incorrect - you need to be actively engaged to take quick evasive/corrective car actions. The old regions of brain only help you to duck if a projectile is coming towards you, not to steer the car or brake quickly.

Ok, that's fine. I just disagree with you. I think you're kidding yourself if you believe we drive vehicles solely with our somatic nervous system-- the autonomic nervous system is doing most of the driving, just like when you ride a bike, or walk and chew gum at the same time. See the backwards bike video from "Smarter Ever Day" for further evidence of that. See how far active engagement takes him when he pits one side of his nervous system against the other. It's a long fight to overtake reflex with conscious desire.

 

[CyberGus]

Since lizards don’t drive, it won’t. I once hit a curb when parking as I was distracted by talking on phone (hands free).

Your move.

 

[drtimhill]

Not we, you. I do know. I'm wondering why you think it would at all be worse to make the driver look forward vs not. You really think AP is safe if you keep your hand on the wheel but close your eyes? (of course you don't)

ok, what you "know" is that, for you, and you alone, you have the opinion that OP appeared to be better than AP. Im not saying you are right or wrong, I'm saying that you cannot extrapolate from your one-off subjection experience to an objective "I know OP is always better than AP for everyone" without some serious unbiased research to back that up.

You keep asking me what I think, but that is as irrelevant as what you think. I have my own opinions, based in different experiences, but I'm not going to say "I'm right and you're wrong" since that is a childish waste of time. What I said is that there is no factual research that I'm aware of on which to base an informed opinion.

Personally, I'm not at all sure any driver alert system is much use. First, I don't see why the car is responsible for making sure the driver is alert in the first place. What next, a built-in sobriety test? Second, we are not that far away from when it will be safer for the car to continue in control when it detects a driver that is not alert anyway. Say someone gets in the car dead drunk, starts the car driving them home, and falls asleep. What should the car do? In your world, it would cancel autonomous driving and hand driving back to a drunk driver so he can cause mayhem.

But of course these are my opinions, and I'm not saying we should do away with monitoring systems just because I say so, I'm saying we should not have knee-jerk "of course we need X" responses before we think things through carefully and rationally.

 

[hellchris]

Say someone gets in the car dead drunk, starts the car driving them home, and falls asleep. What should the car do? In your world, it would cancel autonomous driving and hand driving back to a drunk driver so he can cause mayhem.

I do believe driver monitoring makes for safer highways overall, but I do not support the idea of using these systems to punish the driver. "Autopilot Jail" is dangerous IMO and doesn't make sense especially if you consider the self-driving systems to be a safety system. Why take away a safety system? Would it make sense to take away a driver's seatbelts for the duration of a drive if he/she misuses them once while cruising? Of course not.

Driver monitoring and driver punishing are related but not the same thing, please don't confuse my opinions on them. In your sleeping/drunk driver scenario (this is just an idea) perhaps the car shouldn't return control to the passed-out driver, but instead pull over/safely come to a stop. Hell, I've been put in autopilot jail before for letting the car get above 90 for a split second. No AP for the rest of the drive, and that isn't safe IMO. Better would be to lock AP into a "lawful actions-only mode", lock the cruise to the exact speed limit etc, but don't remove AP completely. It's mostly irresponsible drivers that will be put into autopilot jail, we shouldn't be forcing them to drive manually after their "offense" IMO.

 

[Knightshade]

I do believe driver monitoring makes for safer highways overall, but I do not support the idea of using these systems to punish the driver. "Autopilot Jail" is dangerous IMO and doesn't make sense especially if you consider the self-driving systems to be a safety system. Why take away a safety system?

Because while using it properly is safer than not using it- using it improperly is NOT

 

[DIsaacs]

I'm fine with the cabin camera, but would like access to all that data. I not only want access to the video feed (which should be included with Sentry and Tesla Cam recordings), but should have access to the labels the AI is placing on me and want to know what is being done with that data. If my insurance is going to go up because I was fumbling with my phone at a red light, I should know about it.

I think cabin audio should be recorded as well, but I should be able to toggle off both cabin audio and camera as easily as I can in a video conference call.

Seems like a little piece of tape would do the trick.

 

[AlanSubie4Life]

Couple videos from @verygreen showing the system in action.

General discussion:

https://twitter.com/x/status/1379093474130198529

With detection overlay:

https://twitter.com/x/status/1379928419136339969

Night detection overlay:

https://twitter.com/x/status/1380238569030160400

What say you now, @Knightshade ? I assume (and it's probably correct) you'll say that it's not reliable enough. Though I do wonder whether it is actually pretty good at detecting the most common causes of driver inattention (meaning, an improvement over the torque sensor). The dark performance doesn't seem too good though! And if you slide out the visor you can block the camera...I guess just fallback to torque sensor in that case.

They're using this now, apparently, for driver monitoring, if you agree to enable it. HW3 only though; HW2.5 no support.

 

[Knightshade]

I mean.... the dude fooled the system surprisingly well with a 2D picture taped in front of the camera once he got it in the right spot...

And the field of view of the driver is even smaller than originally thought- you can't even see the top of the steering wheel nor either of the hands below the dash.

IIRC there was also a bit where the camera was about 99% blocked simply by putting the sun visor down.


The night shots were interesting... it seems if you're in a relatively bright urban environment ambient headlights and street lights might be 'good enough' for it to work "most" of the time but still got a decent # of too darks.... and certainly anywhere rural you'd be SOL seeing anything with the IR hardware other dedicated systems use (or wearing sunglasses other than just guessing based on neck position since it lacks HW to see through polarized lenses).


So it's certainly better than nothing.

It MIGHT be "good enough" for the EUs driver monitoring requirements they'll have to have on every new car in about a year- in which case kudos to Tesla engineers for figuring out a way to take something never meant to do this job and getting it sufficient to slip past regulators without adding HW.

But I wouldn't be shocked if it's not.