Light touch, Autopilot nags

[D.E.]

I have a relatively light touch on the steering wheel. I see the “hold the steering wheel” nag fairly often. I give the wheel a little jiggle and the nag goes away.

I've read the whole report from the Florida fatality. In it is an illustration of the enormous amount of data the car constantly records while we drive. It records every touch of the brake pedal, when it senses the hands on the wheel and when it doesn't, along a record of every warning. If anyone ever has cause to do forensics exam on my car, they'll assume distraction or inattention when they see the frequent hands off warnings.

I like Autopilot but I don't quite trust it. I watch it like a hawk. I understand that each release is the latest in a very long series of “betas”, only as good as the caffeine levels in an unnamed, overworked, and under appreciated low level programmer. On the other hand, I am the one and only, an original. In my case, inattention during Autopilot use isn't going to be likely. So if I should meet my demise while using Autopilot in my Model S, please know that I actually was paying attention and my hands actually were on the wheel, no matter what that car data recorder says.

Are others seeing that nag when they really are holding the steering wheel?

 

[alcibiades]

Are others seeing that nag when they really are holding the steering wheel?

Yep. Routinely.

 

[RDoc]

It's my impression that the car wants more feedback since some update about a year ago. I generally hold the wheel near the bottom with the weight of my arm on it with a moderately tight grip and don't see the nags in general, although they're not gone entirely even though I'm always holding the wheel (and watching what the car is up to).

 

[brkaus]

Are others seeing that nag when they really are holding the steering wheel?

Yes, on straight smooth roads.

 

[croman]

Tesla's driver engagement monitoring system is a bad joke. I also watch it like a hawk because I know that it can fail at any second but I get nagged too.

Everyone should be less judgey because the system doesn't actually record driver engagement, just a poor proxy in terms of wheel torque.

 

[P100DHG.]

I have a relatively light touch on the steering wheel. I see the “hold the steering wheel” nag fairly often. I give the wheel a little jiggle and the nag goes away.

I've read the whole report from the Florida fatality. In it is an illustration of the enormous amount of data the car constantly records while we drive. It records every touch of the brake pedal, when it senses the hands on the wheel and when it doesn't, along a record of every warning. If anyone ever has cause to do forensics exam on my car, they'll assume distraction or inattention when they see the frequent hands off warnings.

I like Autopilot but I don't quite trust it. I watch it like a hawk. I understand that each release is the latest in a very long series of “betas”, only as good as the caffeine levels in an unnamed, overworked, and under appreciated low level programmer. On the other hand, I am the one and only, an original. In my case, inattention during Autopilot use isn't going to be likely. So if I should meet my demise while using Autopilot in my Model S, please know that I actually was paying attention and my hands actually were on the wheel, no matter what that car data recorder says.

Are others seeing that nag when they really are holding the steering wheel?

This is my exact situation

 

[luckyj]

Are others seeing that nag when they really are holding the steering wheel?

As far as I know, there isn't a sensor on the steering wheel. That means the car has no idea when you are really "holding" the wheel.

The way I understand it, the steering system senses torque on the wheel. If it doesn't feel a little resistance or slight input from you for a while, it begins to assume you are not "holding" the wheel.

So yes, if you are moving your hand and arm to follow the wheel closely and don't periodically give a slight amount of resistence or input to it....the system thinks you are not holding the wheel. Or for a long straight away, you'll have to intentionally give some slight input every once in a while.

 

[davinci2017]

Yes on freeways I do nudge it a bit every now and then. Its become second nature at this point.

 

[Az_Rael]

Are others seeing that nag when they really are holding the steering wheel?

Yep, quite often. I have even had it escalate a couple of times and I wasn't able to dismiss it despite applying as much torque to the wheel as I could without disengaging. Had to manually disengage with the stalk so I wouldn't be put in the "penalty box" when I actually had my hands on the wheel. It was like it locked up for a bit or something and wasn't reading the inputs from the torque sensor.

And, yes, I have thought about this scenario with Tesla's press releases related to AP crashes.

 

[Economite]

So if I should meet my demise while using Autopilot in my Model S, please know that I actually was paying attention and my hands actually were on the wheel, no matter what that car data recorder says.

Should that happen (g-d forbid), you can be assured that Tesla will announce cherry-picked data in a manner that implies that you were horribly misusing AP. See, eg, the last paragraph of this article.

Tesla Crash In Utah Under Investigation By U.S. Safety Regulators

 

[Snowstorm]

I got used to juggling the wheel every 30 sec or so and rarely get nags now. When I used to just hold the wheel, I get nags on a regular basis. I don’t know what else can be done without extra hardware. I really hope Tesla doesnt nerd this and make it totally useless by some crazy nag scheme every 10 seconds.

 

[D.E.]

Should that happen (g-d forbid), you can be assured that Tesla will announce cherry-picked data in a manner that implies that you were horribly misusing AP. See, eg, the last paragraph of this article.

Tesla Crash In Utah Under Investigation By U.S. Safety Regulators

This is from SOUTH JORDAN, Utah (News4Utah)

• The vehicle registered more than a dozen instances of her hands being off the steering wheel in this drive cycle. On two such occasions, she had her hands off the wheel for more than one minute each time and her hands came back on only after a visual alert was provided. Each time she put her hands back on the wheel, she took them back off the wheel after a few seconds.

The real story should be that this superb vehicle allowed this silly woman to survive this horrific crash, 60 mph into a stopped truck.

Now back to the hands on-off the wheel question.

When my nag screen “visual alert” is provided, the car has already logged 60 seconds of my hands “off the wheel”. It seems the car requires constant pressure against the torque sensing system and any brief interruption in that pressure will be logged as period of “hands off the wheel”.

If Tesla is going to report this sensed lack-of-torque to law enforcement agencies as factual eviidence that the hands were off the wheel, it seems to me that they either need a more accurate sensing system, or Tesla should more accurately report it as periods of “insufficient pressure against the sensing system which may or may not indicate hands off the wheel”. As it is, Tesla is setting us up for lawsuits. Unless one has a camera recording the hands on the steering wheel, it will be difficult to refute Tesla's assertion. Maybe there really is a reason to weight the steering wheel with a water bottle during Autopilot use. At least there wouldn't be those damning log entries. The down side would be the car would steer with the bottle weight should the Autopilot disengage. Also the amount of torque provided by the bottle could vary as the car bounces along the road. A heavy bottle and a pretty good bounce might actually disengage the autosteering.

I wonder how much water I should put in the bottle. I suppose it'll depend on the bottle's distance from the center of the wheel. Maybe water isn't the best option. I could make a weight and a clip. Then I could make it adjustable. I'll have to disguise its purpose, it wouldn't do for an investigator to find me using a device with the sole purpose of defeating one of Tesla's safety monitoring mechanisms. No, the water bottle approach is better. It seems I've got some thinking to do.

 

[kirkbauer]

Agree all around. I don't agree with how Tesla reports this. When I read the description on how much she "wasn't holding the wheel" and how many notifications she received, it sounds like my situation when I AM paying attention to the road and holding the wheel.

In fact, I almost never see the initial warnings to tweak the wheel because my eyes are on the road, not looking at my instrument cluster.

Tesla needs a better system to tell if you are holding the wheel and/or a better system to gently remind you to hold on (eg seat vibrate or HUD).

 

[Doug_G]

The trick is to constantly apply a slight torque to the wheel. I get a lot less nags when I do that. Jiggling occasionally is not very effective.

 

[rborovoy]

Bottom line is everyone in this chain knows that Tesla's autopilot hands-on-wheel sensor is flawed - because it doesn't detect whether our hands are on the wheel, but only whether the driver is applying reverse torque pressure. So, God forbid, you are in court after an accident, and charged with negligence, and a Tesla representative testifies that a "hands on wheel" signal was not detected (prior to a crash), that testimony will be successfully countered by your expert describing how the sensor works, and that in his or her expert opinion, the fact that the sensor didn't detect counter-pressure is no proof whatsoever that the driver's hands were not fully on the wheel, engaged, and ready to act if necessary. So if your testimony is that your hands were on the wheel, your testimony should be more believable. We all need to remember, however, that autopilot is a "beta" system, and merely having your hands on the wheel while autopilot is engaged is no defense to an accident caused by our own negligence, such as daydreaming, or tuning the radio or navigation system, or turning our head to talk to a passenger, or holding a cellphone and talking - all of which would likely be found to be negligent activity and a likely cause of an accident - even though autopilot was engaged and (except for the cellphone example) our hands were on the wheel.

 

[xyeahtony]

I see the nag sometimes, and i just tug the wheel slightly, goes away. not a big deal for me. I'm not trying to drive hands off like some people.

 

[s1rk]

The trick is to constantly apply a slight torque to the wheel. I get a lot less nags when I do that. Jiggling occasionally is not very effective.

I also noticed that applying constant slight torque while holding onto the wheel helps a lot. I was able to use AP for about 40 minutes straight without the nag coming up. Besides, like what tpham07 said, we shouldn't be driving without our hands off the wheel anyways.

 

[D.E.]

I see the nag sometimes, and i just tug the wheel slightly, goes away. not a big deal for me. I'm not trying to drive hands off like some people.

The problem is that when you see the nag, the car has already recorded that you were not touching the steering wheel for the previous 30 seconds. So you were holding the wheel, but when Tesla prints out that record it plainly states that your hands were not on the wheel until the car prompted you to hold the wheel. When you give it that little jiggle the car records that your hand was on the wheel. Then if you don't continue to provide constant counter-torque, the car again starts recording that you are not holding the wheel and if you go 30 seconds, you'll get another nag. But suppose you give it a little jiggle at say 25 seconds, then you won't see the nag, but the car may well have recorded that you were not holding the wheel for that previous 25 seconds.

These printed reports are assumed by all to be documentation that you drive without having your hands on the steering wheel.

So you are right in that it isn't a big deal for you to clear the nag each time, but it may well be a big deal if some lawyer presents one of Tesla's printed logs documenting you weren't driving with your hands on the wheel for most of the trip. He'll also point out that Tesla has plainly told you that you should drive with your hands on the wheel. Since the car's logs “prove” you weren't holding the wheel for long stretches, then you are clearly negligent.

This “driving into a stopped fire engine” case and the Florida fatality both show that the logs are presented as factual records. We know they aren't factual. In a court these logs can still royally screw you. Every time you see that nag there is a damn log entry that documents your driving negligence and the implication is that you only complied with the requirement that you hold the wheel in response to this car's helpful warnings. You can try saying you always drive holding the steering wheel. That lawyer will hold up those logs as proof you are lying.

I suggest people to read the official final report from that Florida accident, not so much to see the tragedy it documents, but rather to see just how detailed and how much information these cars record. And how everything recorded is assumed to be totally accurate. The report is long, the car records are extensive. It is a real eye opener.

 

[rooter]

Too bad no one's thought to set up a poll. I'm certain that the majority hate this frequency of nags.

It's true that the car logs when the popup comes, then when the glow comes, and when the alert comes. And it's also true that Tesla has no other mechanism to try and assure driver alertness, and this is a delicate time with NHTSA for Autopilot. Tesla're doing the best they can to allow us to keep Autopilot.

But the sensing mechanism is not by torque, although torque does make it go away. The optical interrupter on the steering shaft is monitored, and the pinion motor on the steering rack is driven to make the steering shaft be at a specific position on the interrupter. When you tug on the wheel it causes an unexpected difference between the target and the actual steering location and the system sees that as your hand on the wheel.

It's actually monitoring the optical interrupter on the steering shaft above the brake pedal, as a device in the RTOS of the gateway. As root in the gateway if you add some small amount to the interrupter's current value, periodically, that would be interpreted as a tug on the wheel. You wouldn't want to write too large of a difference as that could cause the wheel to turn. What does this mean? Defeating the nag. Of course first you would need to turn off code signing, learn the RTOS, and root the gateway, and I don't know of anyone else who can do a fundamental compromise such as this.

Don't worry, Tesla's known for a long time how things are done, and there is little they can do about it. The only reason for secrets is for commercial purposes of those who have some skills.

 

[luckyj]

Now that is some interesting information, rooter. I've wondered how they were "detecting torque" on the wheel from the driver. It's been somewhat obvious to me that gripping the wheel is not directly sensed. So in my mind, any conclusion via logged data about hands on or off the wheel is bogus. Though over time, it becomes likely that if you are gripping the wheel, there will be some torque detection. But not a certainty.