Wiki MASTER THREAD: Actual FSD Beta downloads and experiences

[Tronguy]

But Tron, you are quite right that the same human factors issues pertain. As a driver, a pilot and a product designer, I too have criticisms of Tesla’s interface designs.

I, too, have Questions about the UI design. My SO also has questions and, in her case, her opinions are worth _infinity_ more than mine, since she's actually had the no-kidding training. Although.. as is true with any college degree, having the BS/MS/PhD shingle doesn't mean that one has the expertise in a particular, narrow field. As one finds out, the higher one's degree gets, the more one finds out that one knows less and less.

If there's a problem with FSD-b's UI, IMHO, it's that Tesla is inventing the UI and the underlying technology at the same time. Yeah, they're trying to make the UI as intuitive as they can. I would say they're doing a decent job at that but, inevitably, they're not going to hit the bulls-eye with every shot on every topic.

I have to wonder, though, what Tesla's UI development group uses for tools in this regard. At one time early in the SO's career, it wasn't unusual for the engineers to literally go out on the street, find random people walking by, and offer them some paltry sum to, "Come inside and do some tests." All on the up-and-up, complete with signed informed consent forms and the (at the time) a $20, non-counterfeit bill. She and her compatriots would then sit on the far side of a one-way mirror counting therbligs while the subject attempted (and sometimes failed) to do some innocuous task. Results of these studies would sometime result in massive redesigns of the office equipment being user-tested.

Thing is, the above general method of running human rats through the maze is kind of the gold standard for UI design: One gets repeatable, testable results that can be, say, published in journals. Yeah, there are, sometimes, variations depending upon who the rats might be. Speaking of aircraft, for example, there might be a real difference in results if the user doing the testing happens to be a no-kidding test pilot, compared to some 70-year-old general aviation type with poor reflexes. Or a newbie with a sparkling new aviation license staight out of flight school. Which is why user selection, doing runs with large numbers of users, and paying strict attention to the outliers, and why there are outliers, is important. I'm just touching on what I've learned from osmosis; it's actually more complicated than that when the Human Factor gets inserted into the control loop.

But, if there's One Thing the SO absolutely, positively hates, it's when some designer-or-other says, "Well, I did the UI this way because it's Obvious!". Yeah. Right. And one doesn't want to know just how many Captain Obviouses there are out there, but there's a lot.

I truly hope that Tesla employs a bunch of no-kidding Human Factors types. Without them there, it's going to take a lot longer to get FSD-b to work correctly with people.

 

[swedge]

I, too, have Questions about the UI design. My SO also has questions and, in her case, her opinions are worth _infinity_ more than mine, since she's actually had the no-kidding training. Although.. as is true with any college degree, having the BS/MS/PhD shingle doesn't mean that one has the expertise in a particular, narrow field. As one finds out, the higher one's degree gets, the more one finds out that one knows less and less.

If there's a problem with FSD-b's UI, IMHO, it's that Tesla is inventing the UI and the underlying technology at the same time. Yeah, they're trying to make the UI as intuitive as they can. I would say they're doing a decent job at that but, inevitably, they're not going to hit the bulls-eye with every shot on every topic.

I have to wonder, though, what Tesla's UI development group uses for tools in this regard. At one time early in the SO's career, it wasn't unusual for the engineers to literally go out on the street, find random people walking by, and offer them some paltry sum to, "Come inside and do some tests." All on the up-and-up, complete with signed informed consent forms and the (at the time) a $20, non-counterfeit bill. She and her compatriots would then sit on the far side of a one-way mirror counting therbligs while the subject attempted (and sometimes failed) to do some innocuous task. Results of these studies would sometime result in massive redesigns of the office equipment being user-tested.

Thing is, the above general method of running human rats through the maze is kind of the gold standard for UI design: One gets repeatable, testable results that can be, say, published in journals. Yeah, there are, sometimes, variations depending upon who the rats might be. Speaking of aircraft, for example, there might be a real difference in results if the user doing the testing happens to be a no-kidding test pilot, compared to some 70-year-old general aviation type with poor reflexes. Or a newbie with a sparkling new aviation license staight out of flight school. Which is why user selection, doing runs with large numbers of users, and paying strict attention to the outliers, and why there are outliers, is important. I'm just touching on what I've learned from osmosis; it's actually more complicated than that when the Human Factor gets inserted into the control loop.

But, if there's One Thing the SO absolutely, positively hates, it's when some designer-or-other says, "Well, I did the UI this way because it's Obvious!". Yeah. Right. And one doesn't want to know just how many Captain Obviouses there are out there, but there's a lot.

I truly hope that Tesla employs a bunch of no-kidding Human Factors types. Without them there, it's going to take a lot longer to get FSD-b to work correctly with people.

Tron, what does your SO think of tesla using steering wheel torque as a driver attention sensor?

 

[Tronguy]

Tron, what does your SO think of tesla using steering wheel torque as a driver attention sensor?

Good question. So I told her this morning that I had taken her name in vain and why. She agreed with me about her hating the Captain Obviousesses of the world. When asked about the driver attention sensors:

1. Knee jerk: She doeesn't pesonally like the torque on the steering wheel approach.
2. Then, after some thought, since she was being asked for a professional opinion: It probably isn't all that bad since it does mean that one is supposed to keep one's hands on the steering wheel. She'd like to think that there might be a better way given time and technology.
3. I pointed out the eyeball detector. She immediately countered with the issues that sunglasses would be an issue with that, with which I agreed. And then I pointed out that some of these eye-trackers have infrared detectors that, presumably, can work with sunglasses, and that I wasn't sure if Tesla had the infrared detectors or not.
4. She then pointed out that the eye-trackers are going to have issues with one-eyed people. There are at least two classes of these:
   1. People with lazy eye, where one eye is actually tracking and the other points in some $RANDOM direction. Sometimes the eye that's "lazy" is consistent, sometimes it moves from one side of the face to the other, and sometimes both eyes "snap to" and track together.
   2. And then there's people who are literally blind in one eye. By accident, by birth, or what-all. We both happen to know one person who had a massive lazy-eye problem from birth; the non-dominant eye on that person was blind. That is, nothing functionally wrong with it, but if the dominant eye closed, it simply got dark for that person. People with a completely missing eye might have a glass eye in there that may or may not move, I suppose.

In either of the above two cases an eye-tracking system used to determine driver attention, from Tesla or some other vendor, may have significant issues with a significant portion of the population. Who might otherwise be able to drive fine, despite the lack of direct depth perception.​

The point of all this is: Eye-tracking systems aren't perfect and have their limitations. In which case, torque sensing on the steering wheel doesn't look all that bad after all.

She then emphasized that, while she is a UI designer, this is not a portion of the field in which she has been directly involved. And would therefore not feel comfortable second-guessing those who actually are involved.

But you asked for her thoughts, and there they are.

Hm. So, a couple of you guys out there are pilots. Pretty sure that Military types have to have Very Good Eyesight, no glasses, and no lazy eye, either, I bet. I'm sure blind people don't get pilot licenses (just like they don't get auto driver's licenses, either); but what if a prospective pilot has lazy-eye, one eye, or, for whatever reason, no depth perception?

 

[old pilot]

Good question. So I told her this morning that I had taken her name in vain and why. She agreed with me about her hating the Captain Obviousesses of the world. When asked about the driver attention sensors:

1. Knee jerk: She doeesn't pesonally like the torque on the steering wheel approach.
2. Then, after some thought, since she was being asked for a professional opinion: It probably isn't all that bad since it does mean that one is supposed to keep one's hands on the steering wheel. She'd like to think that there might be a better way given time and technology.
3. I pointed out the eyeball detector. She immediately countered with the issues that sunglasses would be an issue with that, with which I agreed. And then I pointed out that some of these eye-trackers have infrared detectors that, presumably, can work with sunglasses, and that I wasn't sure if Tesla had the infrared detectors or not.
4. She then pointed out that the eye-trackers are going to have issues with one-eyed people. There are at least two classes of these:
   1. People with lazy eye, where one eye is actually tracking and the other points in some $RANDOM direction. Sometimes the eye that's "lazy" is consistent, sometimes it moves from one side of the face to the other, and sometimes both eyes "snap to" and track together.
   2. And then there's people who are literally blind in one eye. By accident, by birth, or what-all. We both happen to know one person who had a massive lazy-eye problem from birth; the non-dominant eye on that person was blind. That is, nothing functionally wrong with it, but if the dominant eye closed, it simply got dark for that person. People with a completely missing eye might have a glass eye in there that may or may not move, I suppose.

In either of the above two cases an eye-tracking system used to determine driver attention, from Tesla or some other vendor, may have significant issues with a significant portion of the population. Who might otherwise be able to drive fine, despite the lack of direct depth perception.​

The point of all this is: Eye-tracking systems aren't perfect and have their limitations. In which case, torque sensing on the steering wheel doesn't look all that bad after all.

She then emphasized that, while she is a UI designer, this is not a portion of the field in which she has been directly involved. And would therefore not feel comfortable second-guessing those who actually are involved.

But you asked for her thoughts, and there they are.

Hm. So, a couple of you guys out there are pilots. Pretty sure that Military types have to have Very Good Eyesight, no glasses, and no lazy eye, either, I bet. I'm sure blind people don't get pilot licenses (just like they don't get auto driver's licenses, either); but what if a prospective pilot has lazy-eye, one eye, or, for whatever reason, no depth perception?

The "eye" question is one for a FAA medical examiner. They would know the rules governing eye sight and other health issues.

 

[swedge]

Tron, what does your SO think of tesla using steering wheel torque as a driver attention sensor?

Good question. So I told her this morning that I had taken her name in vain and why. She agreed with me about her hating the Captain Obviousesses of the world. When asked about the driver attention sensors:

1. Knee jerk: She doeesn't pesonally like the torque on the steering wheel approach.
2. Then, after some thought, since she was being asked for a professional opinion: It probably isn't all that bad since it does mean that one is supposed to keep one's hands on the steering wheel. She'd like to think that there might be a better way given time and technology.
3. I pointed out the eyeball detector. She immediately countered with the issues that sunglasses would be an issue with that, with which I agreed. And then I pointed out that some of these eye-trackers have infrared detectors that, presumably, can work with sunglasses, and that I wasn't sure if Tesla had the infrared detectors or not.
4. She then pointed out that the eye-trackers are going to have issues with one-eyed people. There are at least two classes of these:
   1. People with lazy eye, where one eye is actually tracking and the other points in some $RANDOM direction. Sometimes the eye that's "lazy" is consistent, sometimes it moves from one side of the face to the other, and sometimes both eyes "snap to" and track together.
   2. And then there's people who are literally blind in one eye. By accident, by birth, or what-all. We both happen to know one person who had a massive lazy-eye problem from birth; the non-dominant eye on that person was blind. That is, nothing functionally wrong with it, but if the dominant eye closed, it simply got dark for that person. People with a completely missing eye might have a glass eye in there that may or may not move, I suppose.

In either of the above two cases an eye-tracking system used to determine driver attention, from Tesla or some other vendor, may have significant issues with a significant portion of the population. Who might otherwise be able to drive fine, despite the lack of direct depth perception.​

The point of all this is: Eye-tracking systems aren't perfect and have their limitations. In which case, torque sensing on the steering wheel doesn't look all that bad after all.

She then emphasized that, while she is a UI designer, this is not a portion of the field in which she has been directly involved. And would therefore not feel comfortable second-guessing those who actually are involved.

But you asked for her thoughts, and there they are.

Hm. So, a couple of you guys out there are pilots. Pretty sure that Military types have to have Very Good Eyesight, no glasses, and no lazy eye, either, I bet. I'm sure blind people don't get pilot licenses (just like they don't get auto driver's licenses, either); but what if a prospective pilot has lazy-eye, one eye, or, for whatever reason, no depth perception?

I'm with her on this, pretty much. Torque sense is not great, but it is not obvious what better a better option would be.

My thoughts:

I have two major criticisms. First is that torque on the Tesla steering wheel is used for three very different functions: 1) steering the car, 2) sensing driver attention, and 3) disengaging auto-steering. While one does torque the wheel to steer, we are sensitive to the position of the steering wheel and not very sensitive to how much torque it takes to turn it. We just apply however much force is needed to turn the wheel as quickly as we need too for the particular situation. The torque required to keep going straight is very, very low, less than is required to show that one is paying attention. And the torque required to disengage is enough to turn the wheel pretty quickly, making the car jerk sideways when it disengages. So the torque to manually drive straight is lower than the torque to prove one is paying attention. And the torque to disengage is more than that needed to make a smooth minor correction. There is not a simple, consistent relationship between the car's behavior and the driver's torque on the steering wheel, and that relationship changes depending on auto-pilot state - off, TACC, or auto-steer.

What really bugs me is that two hands on the wheel tend to balance out the torque, so the car nags completely inappropriately. (Too many hands on the wheel?) In my plane I had an autopilot. When it was engaged, I usually kept one hand on the yoke, but with a very light touch, so I could feel what the autopilot was doing. Since the autopilot is connected through a slip-clutch, the pilot can easily overpower the autopilot if needed. Let go and the AP just slews the yoke to where it thinks it should be after whatever maneuver the pilot make.

I think I would prefer a contact sense for attention, or maybe a more sensitive torque threshold. I think I would also prefer it if auto-steer had a soft connection, so the driver could steer around a problem (pot hole, lane encroachment by another car, lane change, etc) while leaving auto-steer engaged. Perhaps some decent amount of torque required to overpower the auto-steer, but when you stop torquing, autopilot just takes over and smoothly recanters you in the lane. Disengagement by brake or shift lever-up are fine, no need for steering to disengage.

About eye detection, we usually point our nose in the direction we are looking, we don't drive with side-eye. So head angle should be fine to detect sleeping, book reading, extended romantic gazes with the co-pilot, etc.

A problem with contact detection would be gloves, though Tesla could create compatible designer driving gloves like Apple did with watch bands. Another issue might be conductive or pressure applying cheat devices.

But using one signal, driver torque, at three different levels to mean several different things depending on what mode the autopilot is in seems like a recipe for never-ending criticism. I've also accidentally disengaged because of too much friction when AP is turning the wheel, and I didn't notice till we were drifting out of our lane - not safe.

 

[Tronguy]

I'm with her on this, pretty much. Torque sense is not great, but it is not obvious what better a better option would be.

My thoughts:

I have two major criticisms. First is that torque on the Tesla steering wheel is used for three very different functions: 1) steering the car, 2) sensing driver attention, and 3) disengaging auto-steering. While one does torque the wheel to steer, we are sensitive to the position of the steering wheel and not very sensitive to how much torque it takes to turn it. We just apply however much force is needed to turn the wheel as quickly as we need too for the particular situation. The torque required to keep going straight is very, very low, less than is required to show that one is paying attention. And the torque required to disengage is enough to turn the wheel pretty quickly, making the car jerk sideways when it disengages. So the torque to manually drive straight is lower than the torque to prove one is paying attention. And the torque to disengage is more than that needed to make a smooth minor correction. There is not a simple, consistent relationship between the car's behavior and the driver's torque on the steering wheel, and that relationship changes depending on auto-pilot state - off, TACC, or auto-steer.

What really bugs me is that two hands on the wheel tend to balance out the torque, so the car nags completely inappropriately. (Too many hands on the wheel?) In my plane I had an autopilot. When it was engaged, I usually kept one hand on the yoke, but with a very light touch, so I could feel what the autopilot was doing. Since the autopilot is connected through a slip-clutch, the pilot can easily overpower the autopilot if needed. Let go and the AP just slews the yoke to where it thinks it should be after whatever maneuver the pilot make.

I think I would prefer a contact sense for attention, or maybe a more sensitive torque threshold. I think I would also prefer it if auto-steer had a soft connection, so the driver could steer around a problem (pot hole, lane encroachment by another car, lane change, etc) while leaving auto-steer engaged. Perhaps some decent amount of torque required to overpower the auto-steer, but when you stop torquing, autopilot just takes over and smoothly recanters you in the lane. Disengagement by brake or shift lever-up are fine, no need for steering to disengage.

About eye detection, we usually point our nose in the direction we are looking, we don't drive with side-eye. So head angle should be fine to detect sleeping, book reading, extended romantic gazes with the co-pilot, etc.

A problem with contact detection would be gloves, though Tesla could create compatible designer driving gloves like Apple did with watch bands. Another issue might be conductive or pressure applying cheat devices.

But using one signal, driver torque, at three different levels to mean several different things depending on what mode the autopilot is in seems like a recipe for never-ending criticism. I've also accidentally disengaged because of too much friction when AP is turning the wheel, and I didn't notice till we were drifting out of our lane - not safe.

Couple minor snivvies with your excellent reply.

Um. Again, I'm not the HF person: That's the SO. What I do know is that HF types use eye trackers a lot in their research and testing. So, one wants to see how that spanking new UI is being interpreted by a human? Turn on the eye tracker and watch the focus dart all over the place.

And I'm not kidding about the "darting". Even when a person thinks that they're staring fixedly at something, that person's eyes are darting up and down, left and right, and it almost looks like a random walk. It gets even weirder: A year or two ago I discovered that the retinas of humans (and perhaps all mammals) vibrate at somewhere around a 100 Hz or so. This dithering causes the rods and cones to move around minutely and, according to the article I read on the subject, results in better resolution than would be achieved if the eyes just stayed put.

And, something else that I happen to know: As a lens with a focus, ye human eyeball is absolute ***. Even the cheapest lenses in the cheapest cameras one can buy are better than what we carry around on our faces. Astigmatism, lousy response time, you name it. Yeah, we do have high dynamic range that puts CCDs and film to shame, but that's about it.

So, if one thinks about it, the image that the eye is nominally receiving is jerking all over the place, vibrating like mad, is badly distorted, and we haven't even gotten to floaters that, somehow, we manage to ignore. How come we can see at all?

Answer: It's not the hardware, it's the software. The image that one sees is a construct, put together in one's brain, and is (at least) a couple of levels removed from the raw data. The sensorium where what is right in front of one is nice and sharp.. and extends sideways to the things that aren't sharp, but we think they are, because the eye zips its focus point around gathering data that the brain puts into a stable, unified image that kinda appears sharp everywhere. But ain't.

We're also good at noticing changes in things. But that's evolution for you: Those in one's evolutionary history going back to the lizards and before who didn't notice predators in time ended up in the inside of predators, rather than on the outside.

So, eye trackers. They're detecting the data gathering devices on a human that're only roughly under conscious control. They're attempting to figure out what the eye's pointing at in order to get an idea of a driver is Paying Attention to the road. And what's the eye doing? Well, unless the driver is dead, it's jumping its pointing angle all over the place. The people who make eye trackers and can get some kind of data out of them have my respect.

I agree with your comments about the torque sensor needing better definition.. and maybe better hardware. But the more I think about eye trackers, somehow, the less confident I am in them being the end-all of tracking attention.

 

[old pilot]

I use this and never have any nags. I just rest my right hand on it. Any other method I have tried does not work for an extended period of time.

 

[SucreTease]

What really bugs me is that two hands on the wheel tend to balance out the torque, so the car nags completely inappropriately.

The steering motor does not measure whether weight is balanced or not; rather, it detects a resisting counter-torque to the car's own attempts to turn the wheel (or hold it straight). In other words, if, during the car's frequent small steering-wheel adjustments, it finds that the torque required exceeds some threshold value, it presumes that hands on the wheel must be what is causing that.

You could hold the wheel extremely rigidly in place without applying any weight to either side (like forcing your knee against the wheel, which doesn't apply torque itself, but resists turning by friction) and it would still detect your presence during its attempts to make tiny adjustments. Weight differences between the hands is only one way to provide a counter-torque.

 

[swedge]

The steering motor does not measure whether weight is balanced or not; rather, it detects a resisting counter-torque to the car's own attempts to turn the wheel (or hold it straight). In other words, if, during the car's frequent small steering-wheel adjustments, it finds that the torque required exceeds some threshold value, it presumes that hands on the wheel must be what is causing that.

You could hold the wheel extremely rigidly in place without applying any weight to either side (like forcing your knee against the wheel, which doesn't apply torque itself, but resists turning by friction) and it would still detect your presence during its attempts to make tiny adjustments. Weight differences between the hands is only one way to provide a counter-torque.

I am not convinced of your description of how driver torque is sensed. The torque the AP servo applies is comprised of many factors including system friction, actual force required to turn and hold the front wheels in the desired angle, as well as driver input. The force to hold the front wheels depends on speed, truning radius, tire pressure, tire where, road surface, etc. Spinning the wheel faster also requires more torque. So I seriously doubt that ferreting out the driver's contribution using only the servo torque will do the job.

I suspect that instead the sensor is between the steering wheel and the power steering unit. There is a small amount of free-play in that connection, and the wheel seems to center itself in that free-play range. You can feel this by getting in the car without turning it on, i.e. not touching the brake pedal. The wheel is firmly locked, but you can turn it a small amount. So, perhaps the car senses this deviation from the center of the free play? Who knows.

In any case, I find that two hands on the wheel on a straight road, often does not prevent the nag. Neither does one hand with a light touch on the wheel. The failure of the car to detect these is annoying, and requires me to adapt my steering wheel habits to less comfortable techniques. To prevent the nag (and eventual forced disengagement) one must apply a pressure as if turning the car away from the desired course. Very strange, that. So, no pressure tells the car the driver is not paying attention, light pressure tells the autopilot all is OK, and more pressure tells the autopilot to partially disengage. Of course, if AP is not engaged, these same pressures mean completely different things: none means we are going straight and all is OK, light pressure means turn to stay in the lane, more pressure means turn faster.

 

[jebinc]

@Ramphex I turned FSD(j) on today just to remind myself of bad this version is. Yep, good for another month as I got my fix (of junk)!

 

[Ramphex]

View attachment 931658 I use this and never have any nags. I just rest my right hand on it. Any other method I have tried does not work for an extended period of time.

Is that a leather Suicide Knob?

@Ramphex I turned FSD(j) on today just to remind myself of bad this version is. Yep, good for another month as I got my fix (of junk)!

I have yet to drive on the latest revision of junk. I was hoping the 11.4 would be out by the time I got back, but it’s not looking promising. I’ll be home on Thursday and I’ll give it a spin. Can’t wait for it to jerk me around, slam on the breaks, slap me and call me a bastard.

 

[old pilot]

Is that a leather Suicide Knob?

I have yet to drive on the latest revision of junk. I was hoping the 11.4 would be out by the time I got back, but it’s not looking promising. I’ll be home on Thursday and I’ll give it a spin. Can’t wait for it to jerk me around, slam on the breaks, slap me and call me a bastard.

Yes a Suicide knob, works great.

 

[SucreTease]

I am not convinced of your description of how driver torque is sensed. The torque the AP servo applies is comprised of many factors including system friction, actual force required to turn and hold the front wheels in the desired angle, as well as driver input. The force to hold the front wheels depends on speed, truning radius, tire pressure, tire where, road surface, etc. Spinning the wheel faster also requires more torque. So I seriously doubt that ferreting out the driver's contribution using only the servo torque will do the job.

I suspect that instead the sensor is between the steering wheel and the power steering unit. There is a small amount of free-play in that connection, and the wheel seems to center itself in that free-play range. You can feel this by getting in the car without turning it on, i.e. not touching the brake pedal. The wheel is firmly locked, but you can turn it a small amount. So, perhaps the car senses this deviation from the center of the free play? Who knows.

In any case, I find that two hands on the wheel on a straight road, often does not prevent the nag. Neither does one hand with a light touch on the wheel. The failure of the car to detect these is annoying, and requires me to adapt my steering wheel habits to less comfortable techniques. To prevent the nag (and eventual forced disengagement) one must apply a pressure as if turning the car away from the desired course. Very strange, that. So, no pressure tells the car the driver is not paying attention, light pressure tells the autopilot all is OK, and more pressure tells the autopilot to partially disengage. Of course, if AP is not engaged, these same pressures mean completely different things: none means we are going straight and all is OK, light pressure means turn to stay in the lane, more pressure means turn faster.

You are confusing torque (a turning force tangential to the wheel) with "pressure", which is vague and an incorrect usage of the term (force per unit area). Yes, you can apply pressure by squeezing the wheel, but that is irrelevant to this issue.

You are also confusing the amount of torque needing to turn the wheels with the torque needed to turn the steering wheel. These two are nearly unrelated because of power steering; the amount of torque required to turn the steering wheel is far less (by at least a factor of 10) than the torque required to turn the wheels. I know this because my power steering failed (actually had failed to initialize) within a few weeks of my getting the vehicle and it was so incredibly difficult to turn the wheels without the torque multiplication provided by the power steering that I could barely the turn the car around and reenter my driveway.

Because of power steering, the amount of torque required by the steering motor is fairly even (and that level can be customized in Settings), just as the driver doesn't feel that much difference despite all of the factors you cite. The driver's level of torque is an input to the power steering system and it performs the necessary output torque to the wheels.

The only measurements available to the car to detect your hands on the wheel is through the motor which turns the steering wheel and a rotation sensor which monitors how much the wheel has rotated. The computer simply monitors any resistance to its own attempts to steer the wheels. Too much turns off Autosteer; too little is a sign no one is there. There really is no other way for the car to monitor hands on the wheel.

 

[GBMaryland]

The steering motor does not measure whether weight is balanced or not; rather, it detects a resisting counter-torque to the car's own attempts to turn the wheel (or hold it straight). In other words, if, during the car's frequent small steering-wheel adjustments, it finds that the torque required exceeds some threshold value, it presumes that hands on the wheel must be what is causing that.

You could hold the wheel extremely rigidly in place without applying any weight to either side (like forcing your knee against the wheel, which doesn't apply torque itself, but resists turning by friction) and it would still detect your presence during its attempts to make tiny adjustments. Weight differences between the hands is only one way to provide a counter-torque.

One of the most annoying things I find myself dealing with is that the car constantly tells me it's detected a "defeat device" when it's not true.

I actually got mark against me for supposedly having one out of left field... with the car slowing down and alerting all over the place.

(I got mark #2 and I dump truck has flying debris and I stomped on the gas to avoid damage to the car... without being quick enough to disable FSDb....)

The steering wheel hands-on detection is pretty crummy in that it constantly has an issue with my hand being lightly on the wheel.

 

[swedge]

The only measurements available to the car to detect your hands on the wheel is through the motor which turns the steering wheel and a rotation sensor which monitors how much the wheel has rotated. The computer simply monitors any resistance to its own attempts to steer the wheels. Too much turns off Autosteer; too little is a sign no one is there. There really is no other way for the car to monitor hands on the wheel.

Sorry if my terminology was not exactly precise. The rack and pinion steering gear converts steering wheel torque to linear force, so it this context, torque and force are functionally equivalent. As you say, technically pressure is force per unit area, which is different. Though in common use we often interchange the two.

Here is a tear-down video of a model 3 steering rack. link The steering column has what they describe as a torque sensor mounted around the shaft to the pinion, screen shot below. The power steering assist and the autopilot steering are both supplied by the same electric motor, so AP motor torque is not amplified by the power steering as you describe. Interestingly, the motor is connected by a belt to a ball bearing drive to a screw grove machined into the rack. The video does not go into how the steering motor drive power is calculated, or how the steering wheel position is determined. I suspect that in addition to a torque sensor, the device in the photo below also measures the rotational position of the steering wheel.

My point was that the torque applied by the motor could not provide an accurate measure of torque applied to the steering wheel by the driver. This is because the motor has to compensate for both steering wheel and front tire forces. A single sensor could not discriminate between the two.

All of which is off point regarding the human factors issues around using torque on the steering wheel to determine driver attention.

P.S. The torque sensor pulled out of the M3 looked similar to this Bosch unit: link

 

[SucreTease]

My point was that the torque applied by the motor could not provide an accurate measure of torque applied to the steering wheel by the driver. This is because the motor has to compensate for both steering wheel and front tire forces. A single sensor could not discriminate between the two.

The fact that the steering wheel motor can reliably and accurately detect, measure, and translate the driver's relatively small steering input torque into the large output torque needed to steer the wheels necessarily implies that the motor can accurately measure the torque applied by the driver. If it could not, then either the driver-applied input torque, or the output torque to the wheels would vary noticeably.

Instead, the steering experience is entirely consistent and predictable to the driver. This, by itself, negates your assertion that it "could not provide an accurate measure of torque applied to the steering wheel by the driver."

 

[swedge]

The fact that the steering wheel motor can reliably and accurately detect, measure, and translate the driver's relatively small steering input torque into the large output torque needed to steer the wheels necessarily implies that the motor can accurately measure the torque applied by the driver. If it could not, then either the driver-applied input torque, or the output torque to the wheels would vary noticeably.

Instead, the steering experience is entirely consistent and predictable to the driver. This, by itself, negates your assertion that it "could not provide an accurate measure of torque applied to the steering wheel by the driver."

There is no "steering wheel motor". There is only one motor and one separate torque sensor. The torque sensor is on the steering wheel shaft, and the motor is on the steering rack which is mechanically connects to the two front wheels. The torque sensor detects how much torque the steering wheel is applying to the pinion/rack. The motor apply force to the rack via a screw drive by way of a belt. Thus the motor works agains the combined forces from the steering wheel and the front wheels, while the torque sensor measures only the steering wheel torque.

Perhaps when you say "motor" you mean the motor and the torque sensor working together. While they both see the same amount of movement of rack, the torque sensor directly measures the torque applied by the driver to the steering wheel.

Thank you for debating the existence of the torque sensor. Previously I had only deduced that there was a torque sensor but did not know the exact hardware configuration till I look it up for you. Measuring torque on a rotating shaft is a bit tricky, but now I have a clear understanding of how it is done.

Please see the screen shots and watch the video I linked for you before:

 

[slam]

Hi All,

Just wanted to update everyone on my FSD beta-testing. A few days ago I received the 11.4.2 FSD beta and I have to say the improvements are amazing! I previously had 11.3.6 and had to turn it off after a few days.

I have it set to "Agressive".

Lane changes have improved greatly, it's more forcefull and completes more lane changes than before. 11.3.6 would signal and take too long and end up having to cancel. 11.4.2 pretty much does the lane changes 11.3.6 couldn't do. I have had 2 instances where it's had to cancel a lane change, but that was due to people jumping in before the car could start the lane change. 11.4.2 will also accellerate into the lane change.

The handling of the HOV lanes we have here in Ontario has also improved. It no longer seems to want to cross the double lines to get into the regular lanes. That was probably a "learned" bad habbit since so many people do that, which isn't allowed. However, if 11.4.2 sees that the non-HOV lane has room, it'll want to change lanes. I've had to cancel a few lane changes, those currently are the only "interventions" I've had to do regarding HOV lanes.

Merge lane handling still needs a bit of work, but has improved as well. In 11.3.6 it would drive past the end of the merge lane drive on the shoulder and then merge, which is illegal where I am (Ontario). This time it has either stopped or got out of the merge lane before the end. It still doesn't get out of the merge lane at the first opportunity, but if I just hit the signal, it'll attempt to do a lane change earlier.

Turns are also improved. I have HW3, so it will now will creep forward a bit to get a better view and then initiate the turn. Right hand turns are pretty flawless now. Left hand turns have greatly improved, but occasionally there has been some hesitation. I've only had to intervene on two left hand turns in the past 2 days, but that was just due to hesitation. One was a left hand turn onto a curved road, I could see there were no cars, and even the display shows a blue area where the car wanted to go, but it just waited too long. The other time was when I was at an intersection, and the light turned yellow, but after a few seconds 11.4.2 would still not proceed or creep forward (so I intervined to complete the turn). It couldn't see the oncoming lanes because there was a bus had already stopped in the oncomming left hand turn lane and blocked the view.

Biasing away from trucks is less extreme. I don't notice it on straight roads like I had noticed with previous 11.3.* betas.

FSD 11.4.2 has less of a tendancy to get over to the right hand side of the road, much fewer cancelled lane changes. Of course everyone is supposed to go over to the right and leave the left hand lanes clear for passing. But in Ontario, there's no lane discipline. HOV lanes are in the left lane, and I still see this tendancy to want to move over to the right happening.

One quirk have noticed is the yoke/wheel will suddenly jerk one direction and then reverse it. Fast enough it doesn't really change the direction of the car, but I ended up inadvertantly disengaging FSD because my hand put enough torque on the yoke/wheel when it did this jerking motion while turning.

Another quirk I also noticed with disengaging FSD. When you intervine it only disables FSD, the automatic cruise control is still engaged. I think it should drop down completely to manual control. ie. If I've been booting along with FSE at 60 km/h, but intervine to take over a left hand turn. Sometimes in the middle of the turn, the car will start accelerating to 60 km/h pretty agressively, and it ends up being a bit surprising.

So... yeah... wow! Great work and progress by the Tesla FSD team!

Keeping it enabled this time!

 

[Watts_Up]

One quirk have noticed is the yoke/wheel will suddenly jerk one direction and then reverse it. Fast enough it doesn't really change the direction of the car, but I ended up inadvertantly disengaging FSD because my hand put enough torque on the yoke/wheel when it did this jerking motion while turning.

I recently got the new FSD update and I noticed that the car suddenly for no special reason decided to move toward the center of the road.

Luckily I was holding firmly my steering wheel, but my passengers noticed it, and after few occurences I decided not using FSD when I have pasengers.

I'm now waiting hopefully for an update release. I wonder if other users also experienced this issue?

 

[Dutchie]

One other quirk is that the settings always go back to “chill” in the morning or if someone else has driven. The “aggressive” setting isn’t saved in the driver profile.