Auto Pilot Changes already implemented?

[brucet999]

I got the "place hands on the wheel" nag a few times during a 20 mile stretch with AutoPilot on Thanksgiving day. I squeezed the wheel very lightly without any rotation and it went away each time but never ceased steering the car. I didn't time it, and did use auto lane change a couple of times, which would likely have reset any "nag clock", so I can't say how much time between nags.

 

[green1]

For anyone who thinks there's a timed nag. Please prove it. Take a stopwatch, and tell us exactly what the nag time is. It should be reproducible, and the same every time.

 

[Evee]

I don't think it's timed... But there *are* times this system does nag (if even very little) when it's not necessarily having a hard time figuring out what's going on with the road.

 

[LetsGoFast]

For anyone who thinks there's a timed nag. Please prove it. Take a stopwatch, and tell us exactly what the nag time is. It should be reproducible, and the same every time.

There is a stretch of I-95 north of Richmond where I routinely get 40+ minutes without the nag, so its going to be a boring test.

 

[brianman]

I don't think it's timed... But there *are* times this system does nag (if even very little) when it's not necessarily having a hard time figuring out what's going on with the road.

I would rephrase as "when it's not necessarily obvious why it might be having a hard time" -- and then I would agree with you.

This gives the feeling of a time-based nag, um, at times.

 

[green1]

I don't think it's timed... But there *are* times this system does nag (if even very little) when it's not necessarily having a hard time figuring out what's going on with the road.

Just because you can't tell why it's having a hard time doesn't mean it isn't.

Also note that there is one occasion where it isn't having a hard time, but still "nags" and that's curves in the road, on a curve it may be able to see perfectly, but because the camera has a limited field of view and can't see around the curve well, it wants to make sure the driver is awake.

 

[ThreadAbort]

It would be interesting to see if the nag times vary based off Comfort, Standard, or Sport steering modes. Is it harder for it to tell that you haven't touched the steering wheel under certain conditions.

Personally I think it is time based and depending on how much the steering wheel moves. So, a light turn every now and then could reset this timer as it thinks you could have touched the steering wheel.

I would guess that it works by measuring amperage used on the steering wheel motor. It would know that you haven't touched the steering wheel if amperage was pretty much nothing for quite some time. Small turns would cause some amps to flow. Then factor in the confidence level of the line tracking.

I bet if you had the steering mode, confidence level and the amperage readings of the steering motor, and time for last possible steering wheel grab you would be able to figure out exactly when you get the nag.

 

[green1]

Personally I think it is time based

As I mentioned elsewhere. Please prove it. get a stopwatch, and tell us what the time is. it should be reproducible, and consistent.
I'm willing to bet you'll find it to be neither.

 

[Patrick W]

According to the manual if the driver continues to ignore the Hold Steering Wheel warnings:

"Autosteer will eventually decelerate Model S to a full stop and then turn on the hazard warning flashers."

Seems like I read somewhere that before stopping it pulls off to the side of the road but that's not in the manual.

 

[Max*]

According to the manual if the driver continues to ignore the Hold Steering Wheel warnings:

"Autosteer will eventually decelerate Model S to a full stop and then turn on the hazard warning flashers."

Seems like I read somewhere that before stopping it pulls off to the side of the road but that's not in the manual.

Someone posted a video, it does not pull off to the side of the road.

That was the original plan, but I guess it's too risky to initiate a lane change without the user being aware with the cars limited hardware suite.

 

[CuriousG]

There is no time based nag on .106 for me. However there is a nag any time you enter a curve tighter than some specific radius (unknown what radius) as well as any time it loses confidence.
I've done trips where I had to grab the wheel every 30 seconds, and ones where I've gone 20 minutes without touching it. All depends on the road, the lighting, the angle of the sun, the cleanliness of the windshield, the traffic around, etc, etc.

I'm on my 4th day of ownership. On the first day I did try autopilot for a bit on my way back from delivery and didn't notice this issue. Second day again didn't notice any issue although I did get a couple of times where I had to use the accelerator to resume TACC from a complete stop on the freeway when the message was displayed. Then came the firmware update to 2.9.12 which I applied at night. Didn't notice any difference in the 17" screen but come third day of driving with AP on the freeway and it would nag me about keeping my hands on the steering wheel. This happened about 3X during my drive. I keep light pressure on the steering wheel with one hand around the 7 o'clock position. Not sure which grip would acknowledge the hands on steering wheel but I think I tried gripping harder with same position and with 10-2.

 

[Discoducky]

Seems like I read somewhere that before stopping it pulls off to the side of the road but that's not in the manual.

Someone posted a video, it does not pull off to the side of the road.

That was the original plan, but I guess it's too risky to initiate a lane change without the user being aware with the cars limited hardware suite.

Why do you think pulling over to the side of the road was "the original plan"? I'm curious as that was something that I've missed.

Pulling over safely would require many more sensors, improved map data and relevant contextual road segment information IMO

 

[Max*]

Why do you think pulling over to the side of the road was "the original plan"? I'm curious as that was something that I've missed.

Pulling over safely would require many more sensors, improved map data and relevant contextual road segment information IMO

I want to say Elon mentioned it at some point in time, but I could be wrong. Someone at Tesla said that if you pass out or have a heartattack the car will pull over [to the side of the road] safely. I'm not sure if it was an explicit or implicit statement.

 

[Andyw2100]

I want to say Elon mentioned it at some point in time, but I could be wrong. Someone at Tesla said that if you pass out or have a heartattack the car will pull over [to the side of the road] safely. I'm not sure if it was an explicit or implicit statement.

I believe you are correct that someone at Tesla mentioned this. I also think some of the articles published talked about this.

Why do you think pulling over to the side of the road was "the original plan"? I'm curious as that was something that I've missed.

Pulling over safely would require many more sensors, improved map data and relevant contextual road segment information IMO

I'm not sure that's necessarily true, and more importantly, I think I could argue that using the available sensors as follows, it would be safer for the car to attempt to get to the shoulder than to stop dead in the middle of the highway.

I understand that the car can not change lanes on its own with a high degree of safety because it lacks a rear-facing radar, and would be susceptible to cars travelling much faster than it coming up from behind in the lane it was attempting to change into. However the car can see extremely well in front of it. It knows when it has clear road ahead. So, in a situation where the car's systems believe a driver could be incapacitated, and the car needs to safely be brought to a stop, why not, instead of stopping dead on a highway where the car could be rear-ended at catastrophic speed, use the car's excellent acceleration abilities coupled with its abilities to see the road in front of it to get to the shoulder of the road, and then stop? Each time it needed to change lanes it could make sure there was plenty of clearance in front, as well as the 16 feet clearance it can see in the rear, in the lane it wants to change into, and then accelerate quite a bit and make the lane change. The worst case scenario there is that the car cuts someone off, but hopefully since the Tesla would be accelerating significantly, the other driver would be able to react and avoid a collision. That should be easier than avoiding a collision when travelling 65 or 75 MPH and suddenly finding a car stopped cold in your lane. You are expecting cars to change lanes on a highway. You're not expecting them to stop.

I know it seems counter-intuitive for the car to speed up when the driver may be incapacitated, but in this scenario, I really think that with the correct programming and the existing sensor-suite, it would be the significantly safer approach to take.

 

[Discoducky]

According to the manual if the driver continues to ignore the Hold Steering Wheel warnings:

"Autosteer will eventually decelerate Model S to a full stop and then turn on the hazard warning flashers."

Seems like I read somewhere that before stopping it pulls off to the side of the road but that's not in the manual.

Someone posted a video, it does not pull off to the side of the road.

That was the original plan, but I guess it's too risky to initiate a lane change without the user being aware with the cars limited hardware suite.

I want to say Elon mentioned it at some point in time, but I could be wrong. Someone at Tesla said that if you pass out or have a heartattack the car will pull over [to the side of the road] safely. I'm not sure if it was an explicit or implicit statement.

I believe you are correct that someone at Tesla mentioned this. I also think some of the articles published talked about this.




I'm not sure that's necessarily true, and more importantly, I think I could argue that using the available sensors as follows, it would be safer for the car to attempt to get to the shoulder than to stop dead in the middle of the highway.

I understand that the car can not change lanes on its own with a high degree of safety because it lacks a rear-facing radar, and would be susceptible to cars travelling much faster than it coming up from behind in the lane it was attempting to change into. However the car can see extremely well in front of it. It knows when it has clear road ahead. So, in a situation where the car's systems believe a driver could be incapacitated, and the car needs to safely be brought to a stop, why not, instead of stopping dead on a highway where the car could be rear-ended at catastrophic speed, use the car's excellent acceleration abilities coupled with its abilities to see the road in front of it to get to the shoulder of the road, and then stop? Each time it needed to change lanes it could make sure there was plenty of clearance in front, as well as the 16 feet clearance it can see in the rear, in the lane it wants to change into, and then accelerate quite a bit and make the lane change. The worst case scenario there is that the car cuts someone off, but hopefully since the Tesla would be accelerating significantly, the other driver would be able to react and avoid a collision. That should be easier than avoiding a collision when travelling 65 or 75 MPH and suddenly finding a car stopped cold in your lane. You are expecting cars to change lanes on a highway. You're not expecting them to stop.

I know it seems counter-intuitive for the car to speed up when the driver may be incapacitated, but in this scenario, I really think that with the correct programming and the existing sensor-suite, it would be the significantly safer approach to take.

How would the current implementation be able to understand if or where a 'shoulder' is located? This would be dependent on just the camera as the radar and ultrasonics do not have this applicable data set. Current maps do not as well a since they just show basic curvature.

 

[Max*]

How would the current implementation be able to understand if or where a 'shoulder' is located? This would be dependent on just the camera as the radar and ultrasonics do not have this applicable data set. Current maps do not as well a since they just show basic curvature.

I asked the same thing, and joked that if the camera makes a mistake and the shoulder is actually a cliff...

 

[Andyw2100]

How would the current implementation be able to understand if or where a 'shoulder' is located? This would be dependent on just the camera as the radar and ultrasonics do not have this applicable data set. Current maps do not as well a since they just show basic curvature.

I did not realize that the software can't differentiate between the right lane, and the shoulder.

Perhaps this would take some work.

I just can't help but think there really has to be a safer solution than a car stopping dead, perhaps in the left lane, of a highway where traffic is moving at 65-80 MPH.

 

[Chasedrgc1223]

It should be able to... It knows when the lines are solid vs broken.

 

[Max*]

I did not realize that the software can't differentiate between the right lane, and the shoulder.

Perhaps this would take some work.

I just can't help but think there really has to be a safer solution than a car stopping dead, perhaps in the left lane, of a highway where traffic is moving at 65-80 MPH.

It can differentiate due to the solid line. But if I put a solid line in a construction zone, I don't think it has a way of knowing if that solid line is the farmost right lane or if we're in construction or <fill in the blank>

- - - Updated - - -

It should be able to... It knows when the lines are solid vs broken.

Solid line doesn't autonatmically mean the lane over is a shoulder. Also not all roadways have a shoulder, but they do have a solid line. So you can't blindly cross a solid line and assume you're in the shoulder. You might be in a ditch, in a guard rail or over a cliff.

 

[gizmoboy]

I want to say Elon mentioned it at some point in time, but I could be wrong. Someone at Tesla said that if you pass out or have a heartattack the car will pull over [to the side of the road] safely. I'm not sure if it was an explicit or implicit statement.

You're right, he did say it as an example of what a reasonable thing to do would be versus competitors' just disabling TACC and leaving it up to you.