Info on Autopilot + v7

[MarcG]

In the event of not hitting the screen in time then I'd expect the car to switch on the hazard lights and slow to a stop. Fun to speculate.

Bad scenario for Autopilot in a non-jammed freeway, which is the main use case for Autopilot.

Imagine this: you're in the fast lane going 65 mph, traffic flowing fine but fairly busy. You fall asleep or are too distracted to hit the "I'm still with you, please keep driving yourself" button in time. The last thing you'd want is for the car to stop on the freeway! Recipe for disastrous accidents.

- - - Updated - - -

Automatically pulling over is awesome but seems fraught with edge cases (what if I'm on a bridge? No shoulder? Cliffside? Meteor is falling onto the shoulder as I pull over, etc.)

Not possible with current hardware implementation (no rear radar or "proper" rear facing camera). Lane changes won't be fully-automated until adequate hardware is installed.

 

[gizmoboy]

Bad scenario for Autopilot in a non-jammed freeway, which is the main use case for Autopilot.

Imagine this: you're in the fast lane going 65 mph, traffic flowing fine but fairly busy. You fall asleep or are too distracted to hit the "I'm still with you, please keep driving yourself" button in time. The last thing you'd want is for the car to stop on the freeway! Recipe for disastrous accidents.

That's a great point... all of the other TACCs I've used (BMW and Infiniti) will self-disengage if the conditions make it unusable.

If one assumes that a putative inattentive driver is considered another case for unsuitability, then disengaging is about the worst possible response the AP system could have.

 

[Ingineer]

I'm aware of the lack of rear radar and camera, and it's doubtful they could use the existing rear camera, but it may be possible. If incapacitated, I would rather it take "my" chances with a slow right lane change until the car can shoulder itself or find an exit ramp. If it begins signalling and looks for clear sonar, it should be pretty safe for it to slowly execute a lane change. Most drivers have had a human driver ahead of you change into your lane like this when you are going faster, I know I have on many occasions. Usually followed with a bunch of expletives, but no accident.

Stopping in the middle of the lane is super-dangerous, especially in hilly areas and/or with high speeds. Multi-car pile-up waiting to happen.

If it were me programming it, in the event of an unresponsive driver, I would have it first alert Tesla over the cellular link while maintaining the set speed. It would then begin by signalling and slowly getting right, (so slowly drivers will have time to react) while scanning for clear shoulder or an exit ramp, then pull over and stop with the hazard flashers on. Tesla could attempt to call the owner and alert emergency services.

 

[mikeash]

Bad scenario for Autopilot in a non-jammed freeway, which is the main use case for Autopilot.

Imagine this: you're in the fast lane going 65 mph, traffic flowing fine but fairly busy. You fall asleep or are too distracted to hit the "I'm still with you, please keep driving yourself" button in time. The last thing you'd want is for the car to stop on the freeway! Recipe for disastrous accidents.

How about making some loud noises and flashing some interior lights first? If you're just distracted or asleep, that should wake you up. If you fail to respond, something worse has probably happened, and slowing to a stop (with hazards activated) is probably a decent reaction. That's probably a medical emergency, and being stopped with flashers will be better than letting you drive off the road, and ought to get bystanders to investigate.

 

[Ingineer]

That's a great point... all of the other TACCs I've used (BMW and Infiniti) will self-disengage if the conditions make it unusable.

If one assumes that a putative inattentive driver is considered another case for unsuitability, then disengaging is about the worst possible response the AP system could have.

TACC is different from full auto-pilot. Autopilot has a much better chance of lulling the driver to sleep since he isn't engaged. Regardless of the Legal issues, you want to do the safe thing, not the thing more likely to cause an accident.

 

[gizmoboy]

How about making some loud noises and flashing some interior lights first? If you're just distracted or asleep, that should wake you up. If you fail to respond, something worse has probably happened, and slowing to a stop (with hazards activated) is probably a decent reaction. That's probably a medical emergency, and being stopped with flashers will be better than letting you drive off the road, and ought to get bystanders to investigate.

Unless you're just ​in flagrante delicto of an... intimate nature -- in which case you're probably getting a ticket.

 

[MarcG]

Tesla has yet to unleash a secret weapon for Autopilot that will solve all these issues. If the driver doesn't respond to a nag, an electric shock will be sent from a high-voltage harness that was secretly installed in the seat bottom cushion of all Autopilot-enabled models :biggrin:

 

[brianman]

If the driver doesn't respond to a nag, an electric shock will be sent from a high-voltage harness that was secretly installed in the seat bottom cushion of all Autopilot-enabled models :biggrin:

Can I get that if I don't respond to the nag? Sounds like it might be good for my back.

 

[Nevek]

I'm going with TACC stalk. Makes sense to tug it, push it, prod it or whatever once in a while if they need to. My preference would definitely be for a no Nag option! Having to pull the stalk to resume TACC after it's gone into 'HOLD' after 3 seconds is already a bind

This makes sense to me, as pulling the TACC stalk is already the way (other than tapping the accelerator) to signal you are ready to resume cruise after a stop and the timeout. The same idea-"I'm paying attention, and want you to resume/maintain prior programming" would apply. Perhaps it would need the input after some change event being noted, such as extra freeway lanes, traffic cones, or a split freeway.

But I'd still prefer no nag.

 

[Footbag]

Automatically pulling over is awesome but seems fraught with edge cases (what if I'm on a bridge? No shoulder? Cliffside? Meteor is falling onto the shoulder as I pull over, etc.).

I loved the meteor comment!

Interesting comments by everyone, the can on the steering wheel 'hack' is pretty ingenious.

 

[MorrisonHiker]

I hope it doesn't nag too much...but if it does, maybe the entire steering wheel is touch sensitive? Then you wouldn't actually have to turn the wheel but just touch it lightly anywhere.

 

[Todd Burch]

There are a couple decent options here. I like the no nag option, but if a nag is required:

There's no need for the driver to do a specific thing to keep autopilot engaged. Think of a screensaver. ANY input to the computer will turn off the screensaver. I think ANY input could keep autopilot engaged.

1. Any change to ANY setting on the car (volume, window opening, tap on the touchscreen, turn signal, TACC distance adjustment, pedal input, etc.) could reset a nag timer, because that means the driver is conscious. Suppose that nag timer is 3 minutes (a reasonable value IMHO). If you make an adjustment to SOMETHING within every 3 minutes, you'd see no nags at all. That will get you a long way, I think.

2. After three minutes of no control change of any kind, a little message appears on the dash and prompts you if you want to continue autopilot. Again, ANY input to the car resets the timer, since it indicates the driver (or someone else in the car) is conscious.

3. If the message is not responded to within 30 seconds, Music fades out (if playing), a gentle tone sounds, and the car again waits for any input. If there is no input after 15 seconds, a loud, annoying alarm tone plays for perhaps 5 seconds in an attempt to wake the driver. If there is still no input of any kind, one must assume an incapacitated driver, in which case the car should make an attempt to pull over safely. (I think it may be possible to do this with current hardware using hazard lights, the camera, sonar, and very slow lane changes). The car would then prompt Tesla to call you, and if there is no response, contact 911 and send an ambulance to your car.

This way, if a nag is required, it starts out in the least annoying manner and gets progressively more annoying as the concern that the driver is asleep or incapacitated grows.

This would be a good balance to minimize annoyance but still have some sort of confirmation for safety reasons.

You can't eliminate every danger or corner case, but I think these steps would be reasonable and fairly non-invasive.

 

[Andyw2100]

There are a couple decent options here. I like the no nag option, but if a nag is required:

There's no need to the driver to do a specific thing to keep autopilot engaged. Think of a screensaver. ANY input to the computer will turn off the screensaver. I think ANY input could keep autopilot engaged.

1. Any change to ANY setting on the car (volume, window opening, tap on the touchscreen, turn signal, TACC distance adjustment, pedal input, etc.) could reset a nag timer, because that means the driver is conscious. Suppose that nag timer is 3 minutes (a reasonable value IMHO). If you make an adjustment to SOMETHING within every 3 minutes, you'd see no nags at all. That will get you a long way, I think.

While I personally like the idea of "any input", from what I've heard about our cars' functions, I don't think that's likely to happen.

The people that have some insight into how our cars' software is designed seem to be fairly certain that the driving and safety software is completely and totally separate from the other stuff, like the audio player, navigation system, etc. According to them, there are good reasons for this separation. Your plan would require integration of the two separate parts of the software. I don't think that is likely to happen, if it is even possible.

 

[Todd Burch]

Sure it's possible, and I think very doable in fact. It's true the vehicle functions must operate without the other systems (so you can drive while rebooting for instance), but that doesn't mean that the vehicle functions cannot respond to input from, for example, the touchscreen. The touchscreen can control creep, traction control, the parking brakes, and other systems, for instance. No reason to think the autopilot system couldn't respond to inputs from other sources. In fact, I think it's 99% likely that what I described is perfectly doable.

The key is that vehicle systems must operate independently of the other systems. But that doesn't mean there can't be communication between them.

 

[gizmoboy]

...If you make an adjustment to SOMETHING within every 3 minutes, you'd see no nags at all. That will get you a long way, I think.

3 minutes! :scared:

 

[Andyw2100]

Sure it's possible, and I think very doable in fact. It's true the vehicle functions must operate without the other systems (so you can drive while rebooting for instance), but that doesn't mean that the vehicle functions cannot respond to input from, for example, the touchscreen. The touchscreen can control creep, traction control, the parking brakes, and other systems, for instance. No reason to think the autopilot system couldn't respond to inputs from other sources. In fact, I think it's 99% likely that what I described is perfectly doable.

The key is that vehicle systems must operate independently of the other systems. But that doesn't mean there can't be communication between them.

OK, I hope you are correct, because I like that concept.

I probably misunderstood what others had written about the way they believed the software was probably handled.

 

[gizmoboy]

Sure it's possible, and I think very doable in fact. It's true the vehicle functions must operate without the other systems (so you can drive while rebooting for instance), but that doesn't mean that the vehicle functions cannot respond to input from, for example, the touchscreen. The touchscreen can control creep, traction control, the parking brakes, and other systems, for instance. No reason to think the autopilot system couldn't respond to inputs from other sources. In fact, I think it's 99% likely that what I described is perfectly doable.

The key is that vehicle systems must operate independently of the other systems. But that doesn't mean there can't be communication between them.

Thing is, if they want assurance that you're paying attention to the road, having you check media or change temperatures is no assurance (in fact, it's assurance you're not currently paying full attention to the road).

 

[MarcG]

One interesting tidbit from the shareholder call that is currently ongoing: Elon said that the "nag" feature will highly depend on how early access customers will respond to it.

And by the way, he expects the auto-steer and self-parking features of Autopilot to be release to early access customers on August 15th (in 10 days).

 

[TonyOrr]

Yes, so we need to ensure all the beta testers insist on no (or minimal) nag :smile:

 

[Todd Burch]

Thing is, if they want assurance that you're paying attention to the road, having you check media or change temperatures is no assurance (in fact, it's assurance you're not currently paying full attention to the road).

It is not Tesla's responsibility to ensure that you're paying attention to the road--that's the driver's responsibility--and it would be darn near impossible to do that. I don't believe that's their goal at all.

The nags are about ensuring that the driver is awake/conscious.