Video: Here's why you have to keep your eyes on the road when using Autopilot

[S4WRXTTCS]

This is exactly why you DON'T release software like this that is beta. This vehicle could have crashed into the car in the next lane and caused death or severe injury.

To my knowledge every lane-steering system on the market today has places where they don't work correctly, and it will crash into the car next to it if you let it.

A lot of the Adaptive Cruise control systems on the market can suddenly slow down in a curve because the radar gets faked out by the car in the other lane. This sudden braking can cause the person behind you to crash into you.

The AEB systems on the market today can sometimes encounter a false positive that prevents the driver from accelerating. In fact this happened to a Model X and it very nearly caused an accident. It happens on cars by other manufactures too. This kind of flaw doesn't involve anything that's in beta, and is way harder to correct for since it requires an unintuitive action on the part of the driver to correct for it.

Blindspot monitoring on the Tesla doesn't work worth crap and if you rely on it you will crash. Blindspot monitoring isn't in Beta and it was a poorly implemented afterthought.

Expecting a drivers assist package (in any car) to work perfectly is foolish. They all have their glitches, and things they don't deal with very well.

I should add that I'm coming from a perspective of someone who has AP, but rarely use the Lane-Steering element. I rarely use it because of a combination of truck lust (similar issue to the OP's case), and that it takes too much of my situational awareness away. I mostly use TACC on the freeway, and when the road/traffic/weather conditions are warranted I use AP. I don't think the right way is to dumb it down (like MB), but to push forwards. To me we have to push forwards with technology because there are way too many deaths on the roads these day. We need this phase despite the drawbacks to reduce accidents and fatalities on the road. Every incremental step counts, and despite its flaws the Tesla AP is putting a lot of pressure on the automotive industry to quicken the pace.

Over the next 3-6 months the following will happen that will greatly improve things.

An OTA update will enhance the radar and will improve the AEB/TACC

The next generation AP hardware will be released with the TriCamera and this will greatly improve things when it comes to truck lust and the OP's situation.

But, we will go full circle even with those two improvements because then other things will happen and more edge cases will come up (like road debris).

 

[valkeriefire]

I'm not a sound engineer but I tried to increase the volume using some 3rd party tools. The video has a sound that is the same duration and tone pattern change of the Autopilot disengagement tone. Unfortunately it doesn't sound like the autopilot tone, but the length and change in tone are pretty much identical. I believe the reason we can't hear the normal tone itself on the unedited video is because it is drowned out by the sound of the tires going over the white dots since the two events occur almost simultaneously.

 

[S4WRXTTCS]

One question I keep asking myself is, what benefit this current AP gives. This video shows that you have to be very alert all the time, because at any time the AP can divert the car towards other car. I think it is much more stressful than driving yourself.

I definitely feel this way on a 70-80mph freeway full of semi trucks in the middle of the night. Where I don't even bother with AP, and I just use TACC.

With very little traffic on good roads during the day I've never experienced any surprises with AP. It's great to rest a bit during the really boring parts. AP is also really good with stop and go traffic. I do usually take over when I see the stoppage ahead, and then I put it back into AP once I'm crawling along.

Other times I use AP as kind of a backup. I'm not sure how it is in Finland, but in the states people are constantly drifting into someone's lane. It's as if the lanes are just a suggestion. It's partly because of inattention and partly because of crappy roads. Heck some lanes are so bad you have to drive on one side or the other of them. When using it as backup I don't feel any stress.

 

[S4WRXTTCS]

I'm not a sound engineer but I tried to increase the volume using some 3rd party tools. The video has a sound that is the same duration and tone pattern change of the Autopilot disengagement tone. Unfortunately it doesn't sound like the autopilot tone, but the length and change in tone are pretty much identical. I believe the reason we can't hear the normal tone itself on the unedited video is because it is drowned out by the sound of the tires going over the white dots since the two events occur almost simultaneously.

Apparently you're not qualified to post on these here forums unless you're a trained professional sound engineer.

 

[sorka]

There is no mistaking the sound. It's quite loud and can be heard over music of any volume with air conditioning blasting and conversations:

In this video, I have the music set louder than you did in yours. It's recorded with the same BlackVue dashcam you have. The audio has not been modified in anyway.

Your AP was not engaged in the video that you posted. There is no way around this fact.

 

[EVie'sDad]

Several things in this video that aren't ideal for autopilot. I'm speaking from my experience with it:

• Botts dots are being used as the lane markers on the right side. The autopilot is far less accurate tracking Botts dots than painted lane markers.
• The left lane marker is a painted line, not Botts dots. In my experience, AP doesn't like the two lane markers to be different. It's either trying to find Botts dots, or it's trying to find painted markers, never both at the same time. In this instance, I almost guarantee that AP had only acquired one of the lane lines, and I'll bet it was in "painted" mode, thus acquiring only the left line.
• Ever since firmware 2.16.17, yellow painted lines aren't detected as easily as white lines. Further, solid lines aren't detected as easily as dashed lines.
• The AP camera sees lane lines more by contrast than color. The black concrete seam is just as good a contrast against the concrete background as the painted yellow line is, and AP began following it.

In this instance on this particular road, the things that can be done to improve AP:

• Drive in the middle lane vice the left lane so that white Botts dots are on both sides vice a Botts dots/painted line mismatch.
• Follow another car on distance setting 3 or higher, this can improve the lane keeping in marginal conditions.

I wouldn't bother sending this to Tesla. To me, this is very typical AP behavior.

Actually, at highway speeds (50-75 mph),a distance setting below 5 for Car Following may result in unexpected deceleration or intervention required from the driver. I keep mine at 5 and it generally will brake in time for all traffic TACC can detect. I still keep my hands on the wheel and eyes peeled for conditions I don't think the autosteer/TACC can handle.

 

[bhzmark]

+1 on recommendation to use autosteer primarily when following a car in front. That makes it much better.

 

[Garlan Garner]

There is no mistaking the sound. It's quite loud and can be heard over music of any volume with air conditioning blasting and conversations:

In this video, I have the music set louder than you did in yours. It's recorded with the same BlackVue dashcam you have. The audio has not been modified in anyway.

Your AP was not engaged in the video that you posted. There is no way around this fact.

I agree.

 

[valkeriefire]

Sorka your car isn't being driven over lane marker bumbs at the same moment. It isn't an identical test. The lane markers increase the ambient noise in the car at that moment the sound occurs. Nothing I'm going to say will convince you and that is fine.
I still find it shocking you believe I manually drove down the freeway unknowingly through multiple turns. Then I found the exact spot where a lane seam occurs and creates a false lane easily identified by other forum members (coincidence 1). I then unknowingly manually drove in opposite way I would have drifted had I not been in autopilot (logically if the road curves right I would drift left if I thought Autopilot was on but it wasn't, coincidence 2). Then at the moment autopilot should be sounding the disengage tone I drove over the lane bumps (which are loud, coincidence 3) and at that same moment the music just happens to make a tone identical in duration and pattern change to the autopilot disengagement tone (coincidence 4). That is 4 "coincidences" occurring simultaneously. Nothing is going to convince you except a log pull from Tesla but I don't have access to that, I wish I did.

 

[SomeJoe7777]

I'm really beginning to get tired of the snippiness on this forum. I've had to "unwatch" about 4 threads in the past week because I can't read the back-and-forth pointless arguing about minutiae.

So the autopilot disconnect sound isn't present in his video. SO WHAT? Does it matter? I don't think it does. I take him at his word that he had autosteer engaged. The reasons for why autosteer wandered outside the lane markers have been discussed, are not unusual, and are well-documented normal behavior for autosteer. End of story.

The lack of the disconnect sound is easily explained by audio compression algorithms and masking. Other louder sounds like the music and/or driving over the Botts dots cause the psychoacoustic lossy audio compression algorithm to "throw away" masked sounds that would be hard to hear -- that's how they achieve compression. His audio has been compressed once by the BlackVue and then again by YouTube. It's no surprise at all that the audio quality is now lacking.

Again, I take the man at his word and see no further reason to hash around the pointless lack of an audio tone. There is nothing to see here.

Thread unwatched.

 

[sorka]

Sorka your car isn't being driven over lane marker bumbs at the same moment. It isn't an identical test. The lane markers increase the ambient noise in the car at that moment the sound occurs. Nothing I'm going to say will convince you and that is fine.
I still find it shocking you believe I manually drove down the freeway unknowingly through multiple turns. Then I found the exact spot where a lane seam occurs and creates a false lane easily identified by other forum members (coincidence 1). I then unknowingly manually drove in opposite way I would have drifted had I not been in autopilot (logically if the road curves right I would drift left if I thought Autopilot was on but it wasn't, coincidence 2). Then at the moment autopilot should be sounding the disengage tone I drove over the lane bumps (which are loud, coincidence 3) and at that same moment the music just happens to make a tone identical in duration and pattern change to the autopilot disengagement tone (coincidence 4). That is 4 "coincidences" occurring simultaneously. Nothing is going to convince you except a log pull from Tesla but I don't have access to that, I wish I did.

My car is being driving over lane markings. Watch the video. I drive over lane markings several times and if you listen very carefully, you can hear the ba-dump of each one. However, very little of that gets into the cabin and what does doesn't increase ambient noise as these are events spaced apart and in a totally different frequency spectrum plus they are WAY quieter than the interior chime by many decibels.

Anyone who watches and listens to your video and then watches and listens to my video will realize there's no way you wouldn't have been able to hear the chime in your car. It didn't happen.

I'm still willing to give you the benefit of doubt on truthfulness in that maybe you thought AP was engaged when it really wasn't.

I have several suggestions:

1) Post the complete video that includes when you engaged AP.

2) Drive the same stretch or similar stretch and engage auto pilot and then disengage and show us the chime can't be heard in your environment.

3) Same as #2 but start a lane change and then take control manually as the wheels cross over the lane markers.

 

[valkeriefire]

I don't think your test is identical because you're crossing at most a single bump. I'm hitting 8 (2 groups of 4) at 72mph. Which kind of suspension does your car have, air or standard? Mine is standard, which could translate into more noise from the bumps transferred to the cabin. Not sure if that could be a factor but it's possible. On your suggestions 1) I already thought of that and it's the best idea. I'm at work and won't see my Tesla until tomorrow when I get home. I asked my wife to pull the SD card but it may have already been overwritten. If it hasn't I'll post the video from the start. 2&3)I don't think re-doing the route could exactly recreate the scenario either, I'd have to swerve back into the left lane at the exact same moment for the test to be valid, and that's pretty much impossible (not to mention reckless and dangerous). It would be interesting to see if the AP could be faked out again by that seam though.

 

[cinergi]

I, for one, give the OP the benefit of the doubt. Why? Because I've had numerous occasions where AP has done the exact same thing to me. It tried to drive me into a jersey barrier across a very well-defined yellow painted line on the left! Fake or not, it illustrates a *real problem* and *why people must pay attention while AP is active* Period. You cannot fully trust AP. If you do and you're not paying attention and are ready to take the steering over *immediately* then you're asking for it.

 

[Electricfan]

One question I keep asking myself is, what benefit this current AP gives. This video shows that you have to be very alert all the time, because at any time the AP can divert the car towards other car. I think it is much more stressful than driving yourself.

Bingo! That's what I want to know!

Valkeriefire, I have posted two dashcam videos on this forum showing exactly what you showed. Sent both to Tesla. They only responded to one of them, saying its Beta, keep your hands on the wheel, blah blah blah.

I hope Tesla fixes this for all Model S AP gen 1.0 owners, but I have zero confidence they will. I've about decided to sell mine. I wish Tesla all the success in the world, but I've just lost confidence in them, after Elon's childish posts and poor response to J Brown's death.

 

[S4WRXTTCS]

Bingo! That's what I want to know!

Valkeriefire, I have posted two dashcam videos on this forum showing exactly what you showed. Sent both to Tesla. They only responded to one of them, saying its Beta, keep your hands on the wheel, blah blah blah.

I hope Tesla fixes this for all Model S AP gen 1.0 owners, but I have zero confidence they will. I've about decided to sell mine. I wish Tesla all the success in the world, but I've just lost confidence in them, after Elon's childish posts and poor response to J Brown's death.

I'm glad you decided to sell because you have an unrealistic idea of what Level 2 Semiautonomous driving can do. In the beginning you even put others at risk because you didn't understand it's limitations. That wasn't Tesla, Elon's, or anyone else's fault. It was your own ignorance in thinking somehow it would be safe to read a book while the car did the driving.

I would advise you to steer clear of any Level 2 car. There isn't a level 2 car on the market today that will work up to the expectations you have for the Tesla Autopilot system.

Every level 2 system on the market today is a work in progress. They are all really beta in the grand scheme of things.

 

[scottm]

Test cases like this are very repeatable.

The spot on the road is identifiable.

A few more AP drivers can go over the same stretch, and see what your cars do? Go in pairs and one driver can pretend to be a Porsche passing on the right. Hey, why is a Porsche having to pass on the right.. ?? That road is pretty clear of traffic and maybe the Tesla should have been in the right lane anyway... not the point, I know.

Report back.

 

[bob_p]

Lexus (and likely other manufacturers) have a camera trained on the driver's face and can detect when the driver is not watching the road and issue a warning.

Tesla should add this capability until they have hardware and software that is qualified for fully autonomous driving.

With AP mode enabled, the software can issue a warning when the driver isn't looking at the road.

With AP mode disabled, it's also useful to issue a warning when the car is in motion and the driver isn't look at the road.

Relying on the driver to have their hands on the steering wheel isn't as reliable to verify the driver is actually paying attention. Adding the camera would along with warnings when the driver isn't watching the road not only would increase the probability the driver is continuously monitoring AP - it is also a safety feature to detect if someone looks away or closes their eyes when driving on the highway at high speeds.

 

[thimel]

Lexus (and likely other manufacturers) have a camera trained on the driver's face and can detect when the driver is not watching the road and issue a warning.

Tesla should add this capability until they have hardware and software that is qualified for fully autonomous driving.

With AP mode enabled, the software can issue a warning when the driver isn't looking at the road.

With AP mode disabled, it's also useful to issue a warning when the car is in motion and the driver isn't look at the road.

Relying on the driver to have their hands on the steering wheel isn't as reliable to verify the driver is actually paying attention. Adding the camera would along with warnings when the driver isn't watching the road not only would increase the probability the driver is continuously monitoring AP - it is also a safety feature to detect if someone looks away or closes their eyes when driving on the highway at high speeds.

I agree with this. Someone can easily keep their hands on the wheel without watching the road. It is the latter that is important until we get true autonomous driving.

 

[valkeriefire]

I was in Autopilot 100% for sure. Attached is the proof. Good thing I had my wife pull the SDcard on Wednesday because it would have overwritten otherwise. I turned on Autopilot several miles before the incident occured. You can clearly hear the tone although it isn't as loud as Sorka's video. There are several factors that could contribute to this, I have the stock audio and standard suspension. I have no idea what Sorka has but if he has air suspension and Premium Sound then his tone could be louder. My BlackVue mic might also be less effective than his. Road conditions and such make a difference as well. I made a control video which I attached before the incident video, it shows how quiet my Autopilot tone is on a quiet 35mph residential area with no radio.

I have to say I've put way more effort into this stupid argument than should have been necessary. I was sent private messages by several people who told me I should ignore forum trolls who just argue for the sake of arguing. I understand debating for the sake of advancing mutual knowledge, but that wasn't what we've been doing. It was trolling plain and simple. If you don't want to troll, and want to have a constructive conversation, be polite. Sorka I'll give you credit, you tried to be polite, but comments like this didn't come across politely.

Your AP was not engaged in the video that you posted. There is no way around this fact.

This full length video proves it was engaged. The control video proves my AP tone is not as loud as Sorka's is (or my mic doesn't pick it up as well). We can now take the totality of the evidence and say for certain what numerous other people already agreed was fact.

This whole debate has burned me on teslamotorsclub. I've been a frequent poster on numerous other forums with thousands of posts (overclock.net, forum.notebookreivew.com, reddit, etc). I've never been to a forum where thread bashing was so opening accepted as it is here. It's disappointing. I'll wasted so much time proving my case to argumentative people here. I already regret it, I should have done what the people who messaged me said to do, "Just my two cents - you need to ignore the nutjobs on TMC. There are people on here who act like they are married to Elon Musk or are his blood relative, and anything negative about the Model S is a personal attack on them. Therefore, its a lie! Please don't let them bother you!" Another user messaged "Ignore the noise-makers on this forum. They tend to make noise just to hear themselves make noise." From now on, that will be my attitude.

Thank you to everyone who believed me in the first place. To those who didn't here you go....

(note I incorrectly labeled the event as being on 7/20. The BlackVue date of 7/16/2016 is correct)

Hey, why is a Porsche having to pass on the right.. ?? That road is pretty clear of traffic and maybe the Tesla should have been in the right lane anyway... not the point, I know.

Report back.

Actually that's funny you asked, I tried to move into the slow lane earlier (you can hear the turn signal), but the Porche speed up and stayed in my blind spot for awhile and wouldn't let me move over. I could see it in the rear view camera footage, then he passed me in the middle lane, I passed him again, etc. Basically he wouldn't let me over. There was also a Cadillac CTS tailgating him for awhile so there was a good stretch of freeway where I couldn't move over. After awhile I gave up. I was going 74mph in Autopilot, so I wasn't trying to obstruct him. I wish he'd just let me in, I could have avoided this whole debate then.

 

[jmsurpri]

I think all the AP news has a lot of people on edge and on the defensive lately, so hopefully over time it will subside. One datapoint I'll add is that my AP chime is also pretty quiet. I've read posts from people complaining about how loud it got after a certain update, but to me, it has always seemed pretty quiet and barely noticeable with car at hwy speeds and radio on. I too have the base audio system so maybe that is the key variable.