Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Hold Steering Wheel every 20-25 seconds?

This site may earn commission on affiliate links.
I think the problem is that the hands on the wheel notion is the wrong approach. The concept is based on two false assumptions. The first false assumption is that there is a system that can determine whether you are paying attention to the road and traffic. Touching the steering wheel, or applying torque to the steering wheel does not ensure that one is paying attention to the road and traffic. Having a camera detect where your eyes are pointing doesn't ensure it either. Until we can connect directly to your brain and read your thoughts we can't be sure what your mind is doing. The second false assumption is that good drivers are always paying 100% attention to the road and traffic and we all know that isn't true. Sometimes we may be thinking about a date last night or an upcoming date or kids or soccer games or meetings or a million other things. Having a system that constantly nags you ensures that your focus becomes on not being nagged rather than on the road and traffic. It is just another form of distraction like a ringing cell phone. What Tesla is saying, in effect, is that we can't trust our steering system to behave the way it should so we want you to constantly monitor it and take the "ball away" (disengage) when it misbehaves. Far more effective would be to have the system only alert you when it encounters a situation it does not know how to handle and to let you take control through the situation (and learn for the next time) and then resume control once it knows what to do. In those long highway drives this should be quite infrequent. I would think in stop and go it would also be infrequent. Then, on the occasion you get an alert, you know you need to react quickly and take control.

Given the inability to reliably detect hands on the wheel, the rationale for using this as a basis for determining whether an accident was caused by the individual not having their hands on the wheel is completely lost. More meaningful would be the system detects a situation it does not know how to deal with and it alerts the driver in sufficient time for the driver to take action and the driver fails to take action. Action here means forced disengagement via brake, accelerator or steering wheel input. The alert can start out gently but must become more intrusive (louder and with flashing lights) as time zero approaches. Time zero is the last possible moment that the driver could take action before an undesirable event could take place.
 
I get it and I got my update this morning. It definitely is nagging much more. My thoughts are, the car knows how fast you are going and if you are in stop and go traffic, do you really need to be nagged...NOPE! If you are doing 75 down the freeway, I would expect more nags. Here's something I would like to see...Motorcycle Mode! Allow you to adjust how far left or right in your lane so motorcycles can pass when you have autopilot engaged!
 
More meaningful would be the system detects a situation it does not know how to deal with and it alerts the driver in sufficient time for the driver to take action and the driver fails to take action.

Even doing this is a very hard problem, still not solved, and very, very difficult given Tesla's current hardware/sensor suite and overall approach. This is the L3 autonomy problem -- how to know in advance that you're about to be in a situation that you can't handle, before you're actually in that situation, in enough time to alert an inattentive driver and for them to gain enough situational awareness to act appropriately (can take 10s or more).

At 80mph, you will travel 350 meters / 1150 feet in 10 seconds, or almost 4 football fields. What sensor is Tesla going to use to look 4 football fields forward? Keep in mind you might travel over a hill top and around a bend in that distance. What compute power can they bring to bear on this (one fairly weak GPU doesn't cut it)? Even with the sensing solved somehow (maps), how will they predict what events might unfold in those 4 football fields of distance over the next 10s? They are very, very far from being able to do this. This is what you need to move from L2 to L3.

If this were easy, not only Tesla but everybody else would be doing it already. If you think it's easy, go build it and become a billionaire overnight.
 
  • Like
Reactions: croman
Perhaps this update with more nags is preparation for future updates going beyond divided highway driving? I use AP on side streets occasionally mainly just to check out its capabilities. This update seems to have significantly better lane keeping performance, e.g. when cresting hills, following other cars, and managing confusing lane markings.

Also, on the highway, there is no noticeable "truck lust".

As for the nags, keeping your hands loose and ready on the steering wheel is less stressful than driving for hours in rush hour traffic or on a divided highway.
 
Even doing this is a very hard problem, still not solved, and very, very difficult given Tesla's current hardware/sensor suite and overall approach. This is the L3 autonomy problem -- how to know in advance that you're about to be in a situation that you can't handle, before you're actually in that situation, in enough time to alert an inattentive driver and for them to gain enough situational awareness to act appropriately (can take 10s or more).

At 80mph, you will travel 350 meters / 1150 feet in 10 seconds, or almost 4 football fields. What sensor is Tesla going to use to look 4 football fields forward? Keep in mind you might travel over a hill top and around a bend in that distance. What compute power can they bring to bear on this (one fairly weak GPU doesn't cut it)? Even with the sensing solved somehow (maps), how will they predict what events might unfold in those 4 football fields of distance over the next 10s? They are very, very far from being able to do this. This is what you need to move from L2 to L3.

If this were easy, not only Tesla but everybody else would be doing it already. If you think it's easy, go build it and become a billionaire overnight.
Hasn't Waymo stated that they believe incrementalism seems harder than going to straight L5? It's an interesting race to watch.
 
I'm sorry. Do you have a point? Feel free to PM if you want to pursue your bizarre worry. Your suspicion of the press isn't relevant to this thread.

I do have a point. I would like to see supercruise used by people who have lived with it for a while, tried to find loopholes, abuse it, etc. Just like what can be found all over YouTube for Tesla AP. Articles written by journalists who spend a couple hours with the system under controlled circumstances, ideal weather conditions, daytime driving and often with a rep from the manufacturer does NOT give an accurate picture of how good a product is.
 
Last edited:
I do have a point. I would like to see supercruise used by people who have lived with it for a while, tried to find loopholes, abuse it, etc. Just like what can be found all over YouTube for Tesla AP. Articles written by journalists who spend a couple hours with the system under controlled circumstance and often with a rep from the manufacturer does NOT give an accurate picture of how good a product is.
That's great. I look forward to hearing all about your inquiry project!

I was merely responding to the poster who thought super cruise didn't work with sunglasses. It does. Conversation complete.

Now I'm going to watch the WC opener and take a break from all this nagging! ;)
 
I do have a point. I would like to see supercruise used by people who have lived with it for a while, tried to find loopholes, abuse it, etc. Just like what can be found all over YouTube for Tesla AP. Articles written by journalists who spend a couple hours with the system under controlled circumstances, ideal weather conditions, daytime driving and often with a rep from the manufacturer does NOT give an accurate picture of how good a product is.
I was curious about this as well. I couldn't find any videos that aren't from car or tech reviewers. I'm not sure who the typical Cadillac CT6 buyer is though. Perhpas they aren't the kind of people who post Youtube videos.
 
@mortman I don't know where you get your 10 seconds as human reaction time. According to various sources on the Internet, here is just one: "The average reaction time for humans is 0.25 seconds to a visual stimulus, 0.17 for an audio stimulus, and 0.15 seconds for a touch stimulus." Source: backyard brains.com. Here is a game where you can test your driving reaction time and it tries to guess your age based on your reaction time: https://www.justpark.com/creative/reaction-time-test/ Even an 84 year old has less than a 1 sec reaction time. What this information says is the car has to be able to see and communicate it's lack of understanding at a minimum of 1 sec distance or 115' at 80mph (your numbers 1,150' for 10 sec @ 80mph). So what you appear to be saying is that in common situations where a lane line is lost, the system cannot see 115' to 230' ahead (1sec to 2 sec) and this is at 80mph. At 65mph this drops to 95'. I think that once you have had a chance to play the game and look at the data you would agree that a person that needs 10sec to react should not be driving. According to Tesla the radar has a range of 250m. The cameras have the following ranges: Narrow forward: 250m, wide forward: 60m, main forward: 160m, forward side looking 80m, rearward side looking 100m, rear camera 50m and ultrasonics 8m. I contend that given these sensor ranges and the required range to identify a problem, the limiting factor is how long it takes the computers to make a determination. If I had to guess, based on the time lane between what one sees and what is displayed on the screen, it would appear to me that the processors are not fast enough to do the computing needed or the code is so fat that it executes too slowly. Maybe they need the newer Nvidia processors?
 
@mortman I don't know where you get your 10 seconds as human reaction time.

I'm not talking about reaction time. I'm talking about the time required for an inattentive person to become attentive and gain situational awareness, well enough to drive the car. This is what's required for level 3 autonomy. L3 means "you don't have to pay attention, go bury your nose in your phone, or watch a movie, or read a book; the car will let you know when you need to pay attention".

I bring this up specifically because you were saying suggesting that the car should just alert you far enough in advance of encountering a problem that you can gracefully take over, rather than nag constantly even when everything is OK. (At least that's what I understood you to be saying; apologies if that was a misinterpretation.) That is pretty much the definition of an L3 system. And it's hard.

Sorry, I don't have references on the "10 seconds" figure handy, but it's highly variable anyway. It can take much longer than 10s to gain situational awareness in complex environments, or much less when things are more straightforward.
 
I could buy 1 to 2 sec but 10 sec is an eternity and not a realistic number. You would have to be in a sound state of sleep to take that long. Count out 1001, 1002,... to 1010 while driving and you will see that it is way too long a time. I could see people over reacting, i.e. hitting the brakes in a panic on a alert but that can happen with normal driving without EAP when an unexpected situation arises.
 
  • Like
Reactions: JohnnyG
This has not been an issue for me - I rest one hand on the bottom of the steering wheel and never see an alert. Honestly, after the latest updates, the auto-steer has been fabulous in my AP1 Model S.

Defensive driving courses teach that both hands should be on the wheel at all times (apart from brief moments to use indicators, controls etc.)
A sudden event such as a blown tire, pothole, or Autopilot glitch requires a firm grip on the wheel to save the situation.
One hand at the bottom of the wheel = dangerous driving

 
  • Like
Reactions: rnortman