Where is the enhanced autopilot?

[TaoJones]

I remain unconvinced that AP2 Autosteer is even at parity with AP1.

Note as well the gift of phantom braking that AP2 has bestowed upon us. It's not just overpasses. It's adjacent traffic and a small number of other scenarios. Still. For well over 1.5 years now.

With regard to the lane changing for certain roads - sounds as if the road tagging is different/updated, without any improvement per se to AP2. Similarly if they ever fix the egregious number of speed limit errors, then there would be fewer instances of tickets or tailgating (read that, lower risk), but this would be an effect of database errors being fixed - not having anything directly to do with AP2's functionality - enhanced or, heh, reduced.

 

[croman]

That hypothesis was absolutely true before firmware 10.4. But not true after firmware 10.4. After 10.4 the issue is now speed. I personally drive on surface streets with AP all the time at < 50 and this condition happens frequently st stop lights and it never fails. Do you not drive under the same conditions I am talking about? Very easy to see the difference. I have read many posts of others that agree with me.

This is absolutely true. Anything untracked under 50mph is recognized and creates a braking action even before it is displayed on IC. Never had an issue since 10.4 though 2018.18 had given me some pause and had me manually override on curves when lead car departs. Still, it has handled that well in most cases.

I'm not sure why greater than 50mph is an issue for the system but before 10.4 it could do up to 30mph just fine, so there is iterative Improvement as the system's vision improves. Maybe the system is so near sighted it can't verify the radar return with vision to confirm braking above 50mph.

 

[barjohn]

@d21mike, I have driven in Torrance on the surface streets (my daughter lives in Hermosa Beach) and I am surprised you feel secure using AP in that area. Traffic can be very heavy and somewhat crazy near the beach communities of Southern Cal. People cut in frequently. I was picturing you living in some rural community with treelined streets and no traffic. Yes the speed limit is under 50mph in most of those areas but many people drive over 50 much of the time. By the way doing the test the way you described it would prove only that I would chicken out before anything could happen whereas doing it the way I described would require ultimate faith in its infallibility. If in fact you are doing this every day in your daily driving you are in effect doing the test thus it should be easy to replicate. I am guessing that you don't come up on a stopped car you didn't see before it stopped often enough to be certain. I am confident enough in most stop and go driving to trust it to a degree but I frequently drive the 55 FWY east fro Newport Beach to Riverside and anytime after 2PM it is a zoo. Cars are constantly changing lanes and trying to jockey for a position one car up and pulling in front of you with inches to spare happens too often for me to trust EAP given its lag in realizing what just happened before it takes action.

 

[sorka]

Supposedly lane changes on surface streets has been activated in certain areas (there are reports from NL and DE, for "highway like" surface streets; "Supposedly" because I couldn't verify it myself yet as I'm in SEA atm).
Sign recognition is still notably absent. As is "all vehicles" rendering in the IC, so AP1 feature parity is still absent 1.5 years after it was supposed to arrive, even if Autosteer has somewhat surpassed that of AP1 by now.

Um, you misunderstand.

That's the $5K for enhanced auto pilot. I'm talking about the extra $3K fort this:

 

[tragamin]

I now own a MS since december 2016, I bought the car because of the announcement of AP V2 with self driving feature. I-m now on FW 2018.20.1 and i can say that driving now feels like 12 month-s back. It-s really annoyoing. The only thing i love is the smooth driving without gear switching and the acceleration. The rest is not really working and some things are really dangerous.

 

[pp61]

I agree, the design of the body and the smooth driving is the main reason for me too. Nobody would buy a car with so called features/improvements that are not activated.....
A friend asked which Tesla S to get, I adviced getting an AP1, had one 3 years ago it drives the same and better AP features.
Clearly the energy of Tesla is not focused on activating (promised) features in the S but to make more 3, get the Y etc.
With the current attitude I think the majority of the Tesla drivers will move in 2020-2022 to electric cars from manufacturers that deliver FULLY functional cars.

 

[CK_Stuggi]

Um, you misunderstand.

That's the $5K for enhanced auto pilot. I'm talking about the extra $3K fort this:

What exactly do I misunderstand? All I'm saying is there's not even AP1 feature parity 1.5 years in, not to speak of EAP (features that only AP2 can do but not AP1) or the slightest glimpse of FSD.

So... What am I missing?

 

[diplomat33]

I think it is very possible that Tesla has decided to forget about the "enhanced" features of autopilot for now and instead, is focusing on finishing FSD as soon as possible.

 

[GeorgeSymonds]

I think it is very possible that Tesla has decided to forget about the "enhanced" features of autopilot for now and instead, is focusing on finishing FSD as soon as possible.

I think its quite the opposite. FSD won't just appear - it would be suicide without many very visible miles driven under FSD - just look at what others are doing.

I suspect there will be a constant stream of enhancements, some EAP and some FSD. As an idea, I think they'll include (and some of these have been mentioned by Tesla). Each will have its own little NN and logic

- Sign detection, especially Stop signs and speed limits
- And then obeying the above at a later time (give time for confidence to build)
- Traffic light detection and then obeying (much harder that static signs)
- Enhanced blind spot recognition and intervention (ie judging closing speed of cars from behind)
- Auto lane change
- Highway junction navigation

Many of these could include a "detection" phase, a "use" stage and an "exploit" stage - ie warning over an approaching stop sign could be a safety feature and all HW2 cars get it. a simple stopping at a stop sign is EAP, but pulling away again could be "FSD" (may not be the best example).

You build up the library of these features while all under the basis that the driver is still in control. Tesla have never said to my knowledge what they really mean by FSD - the "car driving to pick you up wherever" implies the car ican be driverless and therefore in control, whereas the FSD description says "the system is designed to be able to conduct short and long distance trips with no action required by the person in the driver’s seat." which may leave the driver in control. The only area where the driver is not present is the advanced parking feature.

 

[diplomat33]

I think its quite the opposite. FSD won't just appear - it would be suicide without many very visible miles driven under FSD - just look at what others are doing.

Yes, I agree it would be suicide for Tesla to release a "half-baked" FSD. So, no, I don't think that Tesla will just suddenly release FSD all at once when it is not ready yet. Clearly, if Tesla did ever release FSD all at once, it would only be when they have a complete FSD that works.

I suspect there will be a constant stream of enhancements, some EAP and some FSD. As an idea, I think they'll include (and some of these have been mentioned by Tesla). Each will have its own little NN and logic

- Sign detection, especially Stop signs and speed limits
- And then obeying the above at a later time (give time for confidence to build)
- Traffic light detection and then obeying (much harder that static signs)
- Enhanced blind spot recognition and intervention (ie judging closing speed of cars from behind)
- Auto lane change
- Highway junction navigation

Releasing small features incrementally would certainly be a logical approach. But if you are right , based on the current speed of progress with autopilot, it will be 2050 before we actually get FSD! LOL. I want FSD before then.

Maybe it is just wishful thinking on my part but considering that it has been over 2 years since AP2 was announced, I have to believe that Tesla has made some progress towards FSD that we don't know about yet. I doubt that they are still stuck on fine tuning auto steer and have not even begun working on these other features that you mention. So I am hopeful that Tesla is already working on these features you list and might be able to release a L4 FSD in a couple years.

You build up the library of these features while all under the basis that the driver is still in control. Tesla have never said to my knowledge what they really mean by FSD - the "car driving to pick you up wherever" implies the car ican be driverless and therefore in control, whereas the FSD description says "the system is designed to be able to conduct short and long distance trips with no action required by the person in the driver’s seat." which may leave the driver in control. The only area where the driver is not present is the advanced parking feature.

I do think that Tesla will release FSD at L4 autonomy first where the driver is still required to be in the driver's seat. Then later, we will get L5 autonomy.

 

[d21mike]

@d21mike, I have driven in Torrance on the surface streets (my daughter lives in Hermosa Beach) and I am surprised you feel secure using AP in that area. Traffic can be very heavy and somewhat crazy near the beach communities of Southern Cal. People cut in frequently. I was picturing you living in some rural community with treelined streets and no traffic. Yes the speed limit is under 50mph in most of those areas but many people drive over 50 much of the time. By the way doing the test the way you described it would prove only that I would chicken out before anything could happen whereas doing it the way I described would require ultimate faith in its infallibility. If in fact you are doing this every day in your daily driving you are in effect doing the test thus it should be easy to replicate. I am guessing that you don't come up on a stopped car you didn't see before it stopped often enough to be certain. I am confident enough in most stop and go driving to trust it to a degree but I frequently drive the 55 FWY east fro Newport Beach to Riverside and anytime after 2PM it is a zoo. Cars are constantly changing lanes and trying to jockey for a position one car up and pulling in front of you with inches to spare happens too often for me to trust EAP given its lag in realizing what just happened before it takes action.

Please read @croman post just before yours for another person like me that says 2018.10.4 does what I say it does. You do not need to simply take my work for it.

No. Your guess is wrong. Which I said in multiple posts. The car was not "tracked" before. I do understand the difference and it failed pretty much every time before 2018.10.4 unless it was already "tracking" the car. After 10.4 it does not need to track the car before if < 50 mph (mostly for me it is around 40-45).

Regarding Torrance. I live about 1 mile from the beach but I do not drive the beach communities of (Hermosa or Manhattan Beach) that often. Especially where the streets are very narrow. Most of the streets I use AP on are like Hawthorne, Sepulveda (mostly south of Torrance Blvd). I do a lot of testing on Torrance Blvd while going to my favorite restaurant. Most of the roads I drive are at least 2 lanes going each way. Hawthorne and Chrenshaw have 3 lanes each and divided. On those roads if you drive the center lane it is actually quite boring and never an issue. My point is I really understand these roads very well and I know every spot and lane that may cause an issue and I am very careful. I do on occasion drive the 110 and 405 freeways (on the weekend but still can be crowed but not stop and go) to test out auto lane change and to make the transition from 110 to 405 (going north) and early on it would never SLOW DOWN but now (not sure which version) it started to slow in to the transition which is 45 mph.

I do work from my home so my freeway driving is limited compared to others who commute daily. So, if I did not use AP on Surface Streets I would seldom use it. So, more of a hobby/fun for me. But, I have done a few long trips where I used AP most of the way. But again, I avoid SC Freeways during commute traffic times of day.

 

[barjohn]

They could release a small improvement in what I call Waze confirm mode. It sees what it thinks is a stop sign and pops up stops sign! You click the left wheel for thumbs down and right wheel of thumbs up. It gradually learns from its mistakes. When Tesla engineers have enough data showing the cars recognition is 99.999% right it is ready to proceed to the next feature. In this way the cars are being taught how to interpret visual information by tens of thousands of Tesla drivers and improvement should come rapidly. Same for speed limit signs and traffic lights.

 

[diplomat33]

They could release a small improvement in what I call Waze confirm mode. It sees what it thinks is a stop sign and pops up stops sign! You click the left wheel for thumbs down and right wheel of thumbs up. It gradually learns from its mistakes. When Tesla engineers have enough data showing the cars recognition is 99.999% right it is ready to proceed to the next feature. In this way the cars are being taught how to interpret visual information by tens of thousands of Tesla drivers and improvement should come rapidly. Same for speed limit signs and traffic lights.

Isn't that what Tesla is already doing? Tesla has been collecting videos from Tesla cars, including pics of stop signs and using it to teach the neural net.

 

[d21mike]

They could release a small improvement in what I call Waze confirm mode. It sees what it thinks is a stop sign and pops up stops sign! You click the left wheel for thumbs down and right wheel of thumbs up. It gradually learns from its mistakes. When Tesla engineers have enough data showing the cars recognition is 99.999% right it is ready to proceed to the next feature. In this way the cars are being taught how to interpret visual information by tens of thousands of Tesla drivers and improvement should come rapidly. Same for speed limit signs and traffic lights.

I agree. With all of my surface street driving on AP I would love for them to put up a STOP Sign (like in parallel with the speed limit sign to the right). Not sure I want to enter commands to help say the sign is correct. The fact that I actually STOP for it should confirm it is correct. In any case it would make me feel good that they are getting close to recognizing them. I think they should do the same for street lights. Same place on the IC that they would have put the STOP Sign.

 

[The Duke]

I agree, the design of the body and the smooth driving is the main reason for me too. Nobody would buy a car with so called features/improvements that are not activated.....
A friend asked which Tesla S to get, I adviced getting an AP1, had one 3 years ago it drives the same and better AP features.
Clearly the energy of Tesla is not focused on activating (promised) features in the S but to make more 3, get the Y etc.
With the current attitude I think the majority of the Tesla drivers will move in 2020-2022 to electric cars from manufacturers that deliver FULLY functional cars.

I LOL at this, did you not compare the Volvo, MB, Cadillac and BMW offerings when you bought your Tesla? All of them either hype their system or give reasons why it does not work. Granted when you read the fine print disclaimers they are all reasonably accurate. But each dealer I went to said "better than Tesla" except BMW. Is BMW the only "fully functional car" because they are more honest?

 

[GeorgeSymonds]

I agree. With all of my surface street driving on AP I would love for them to put up a STOP Sign (like in parallel with the speed limit sign to the right). Not sure I want to enter commands to help say the sign is correct. The fact that I actually STOP for it should confirm it is correct. In any case it would make me feel good that they are getting close to recognizing them. I think they should do the same for street lights. Same place on the IC that they would have put the STOP Sign.

I agree- its why I think we'll go through stages of evolution - initially it thinks it sees a stop sign and maybe warns the driver - only needs to be 90% accurate to start with like reading speed limits on AP1 cars can pass a few signs before it sees it, but the false positives are ignores and the missed ones it probably never knows about (as we stop for other reasons than signs). Once that has matured introduce is at part of EAP - car actually slows down when it sees a stop sign. Accuracy needs to be much higher to prevent phantom braking. Once thats all working, car stops and they can work on automatic take off - thats much riskier - car determining to go from zero to any speed if harder here than with TACC where its maintaining a distance tot he car in front - it won't have a guide.

The different features can be released together, but many of the EAP features you'd want to see are prerequisites for FSD.

 

[barjohn]

If it did nothing else it would show us that they are working on the feature and that progress is being made. It is one thing to pay for an incomplete feature that is steadily (not unpredictably where the user has to try and figure out what has changed) improving and the steady pace of improvement is visible. Notice in the video on FSD posted earlier the images from the other cameras. Why not allow us to see the images from the other cameras. They are B&W so what. I would love to be able to see the rear side views shown in that video to check for vehicles in my blind spot.

I think Audi is on the right track to do away with the side mirrors. However, the lens needs to be a normal focal length (normal means it presents an image as the human eye would see it) equivalent to 50mm on a 35mm film plane. Too wide an angle and the vehicle appears farther away than it is and too narrow an angel and it appears closer than it is. We judge distance by the size of the image against our ingrained expectations of distance for size.

 

[S100D]

EAP is not released. AP2 is only just now at parity with AP1, there are exactly zero "Enhanced" features available.

AP2 is not on parity with AP1. AP1 reads speed limit signs and slows down and speeds up. It also will do auto lane changes on secondary roads.

 

[CK_Stuggi]

I think Audi is on the right track to do away with the side mirrors.

Forget the camera-only rear view. Tesla planned to do it for the Model X but it's not approved by any regulatory body. Anything like that is and stays a concept until cameras instead of mirrors are approved by regulators. Wouldn't expect that anytime soon.

 

[barjohn]

Have both until the regulators approve it. My Bolt has a rear view display in the rear view mirror and when active there is no inside rear view mirror other than the camera display in the screen built into the rear view mirror. If there is a regulation against this they seem to have gotten around it. Works great when you have stuff in the back that would block the rear view mirror.