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

FIRMWARE UPDATE! AP2 Local road driving...and holy crap

This site may earn commission on affiliate links.
This is pretty much how I would expect it to work.

With AP1 and MobileEye it didn't really work that much better. Any time I went through any major intersection it would lose the lines, and then start searching. It was pretty much unusable on surface streets, and plus it's not entirely useful either except in special cases (like massive traffic, etc). It's not like it can read the stop sign or see a 90 degree turn.

FWIW, in recent 8.0 builds, AP1 has gotten amazingly better at this. It really all started at around 2.52.22, so if you left AP1 before then, I would agree with your statement that it didn't use to do well.

But I've had multiple cases even with no lead cars, the car went through an intersection, all lane lines disappeared on the screen for a brief second but there's still a shadowed outline of what the car thinks is the lane, and it smoothly transitions to the resumed lane lines on the other end. No crazy sudden swerving.

On a similar note, 2.52 also improved AP1's ping-ponging behavior when the car is alternating between detecting the left vs right edge of the lane. Overall, the heuristic for where a lane should be has dramatically improved in recent 8.0 builds on AP1.
 
First, let me start of by saying, I work in technology. As a Cloud Architect, I am heavily involved and very familiar with rapid software development, software development practices etc. I only say this to frame that I would say I am more tolerant than most with this AP2 process. One of the reasons I went with Tesla (first time owner, Dec/2016) was AP2. I fully knew AP2 was not complete, I knew there would be some issues, especially with Tesla ditching the Mobile Eye system and developing their own. I knew it would not be as smooth as the Sales Rep advised me it would be. I was actually excited to be a part of the process. Seeing the car grow and mature into an "autonomous" vehicle, really the first of it's kind, was very exciting prospect. And I knew buying it, that despite what Elon says, the car will never be Level 5 autonomous. I am, or at least was, expecting between level 3-4 autonomy, the car could take drive almost all freeways, exit, drive streets, stop signs etc. I would have to take over in parking lots, etc...Now, i don't even see that happening, at least not any time soon. I fully regret buying FSD. I think AP2 is/will be a worthy investment, they will eventually get there, but i just don't see anything beyond a fully functional and mostly reliable AP2. I just don't understand where the disconnect is between what they showed us in the FSD videos and what we have now. Totally understand that is internal code they are using, but they were showing street driving at 35 mph and it's like not even the basic code logic is being used in what we have. Autosteer AP code is not totally unqiue compared to FSD code. Yes the FSD code will take into account MANY more things, but the base of the code should be the same. We have nothing even close to what they have shown based on my experience below.

Was very excited today to get Firmware (17.5.36).... until I tested it.

imagine you go to the bar, you had 6 double shots, and threw back 5 or 6 beers. Then you decide to be an idiot and drive. That's how the car drives with "Local road driving" AP2. It's basically not usable.

Observations

1. Going through an intersection.. Yeah.. don't do that... Car stars to veer right heavily trying to find a right lane marker

2. Streets with right exit lanes...Yeah... don't do that... It will start to take that right turn lane then suddenly jerk back over into the lane you were in.

3. Road with a smooth 90 turn ahead? Yeah... don't do that... car won't decelerate or even take the turn... it sure does try though.

4. Next to a marked bike lane? Yeah... don't do that... seems to swerve into them.

5. General lane keeps overall just unpredictable


I'd be fine if the update read this way (which it should)

"Autosteer while on Local Roads, only use in this condition:

1. You are in the center lane. Left lanes with left turn lanes and right lanes with right turn lanes do not work.
2. The road has no major curves or turns

Generally Autosteer will only work going straight in a clearly marked center lane."

Something to that effect.

I hope Elon proves me wrong. He has in the past. I just don't see it.

Feel exactly as you do - love "growing up" with the car. Not quite understood the thought process of people buying FSD ( unless you knew you where funding a kickstarter style development)

Thanks for posting the release notes. Really surprised that it has absolutely no warning there for first time users! I guess EAP will never have stop sign and street light detection so on surface streets this is really only useful in heavy traffic .
 
I'm hoping your car is a fluke. However your report doesn't sound encouraging. Sounds like software guys are feeling pressure from above to get out releases. I'm smelling a lot of car buybacks if this doesn't get sorted out soon. I also don't understand why the car worked in the FSD video but now the code somehow got erased out of their servers? Or could it be that the FSD video was fake?
 
I'm hoping your car is a fluke. However your report doesn't sound encouraging. Sounds like software guys are feeling pressure from above to get out releases. I'm smelling a lot of car buybacks if this doesn't get sorted out soon. I also don't understand why the car worked in the FSD video but now the code somehow got erased out of their servers? Or could it be that the FSD video was fake?

Was the FSD video using the new Tesla Vision system or was that car still using MobileEye?
 
Was the FSD video using the new Tesla Vision system or was that car still using MobileEye?

Tesla vision, but it looks like a different codebase. If you see the instrument cluster, there's no cars displayed whatsoever. The official documents Tesla filed to the CA DMV shows like less than 1500 miles of total autonomous driving, and at least an intervention every mile or two, which leads me to suspect it's classic Demo Ware — we are watching the one shot where it just happened to work and not the other ones where it tried to go into a ditch or run over the two ladies jogging, etc etc etc.
 
  • Informative
Reactions: davidc18 and oktane
I'm curious as well but maybe this will give you some hope...
jump to 3:20 if you want to go straight to the best part. Now keep in mind this was posted over a year ago! I think the hardware 2.0 cars will get there in due time. I'm guessing the developers are having a hard time with thousands of different calibrations, at least I hope.

In the video they pass an exit on the right, no veering. I know it's not local streets but if you search around you will find plenty of other videos from Nvidia about the computer and system in our cars. It's just a matter of time.
 
On local roads, our AP1 car slows for turns, easily handles 90 degree bends staying perfectly in the lane

I have just the opposite experience. My AP2 car is in the shop and I got a 2015 AP1 P85D loaner. Drove it for 2 weeks for a business trip in NJ. I wanted to experience as much AP1 as possible so I enabled it every chance I got. The few most dangerous incidents were all at more than 45 degree bends. It literally going straight to the other side of the road. Tried a few more times everyday at the same spot and hope it would learn but it behaved exactly the same. It was a quiet road and hardly have any cars. Otherwise I wouldn't dare to try that again.
 
  • Informative
  • Like
Reactions: Ulmo and malcolm
I have driven about 3-4 times for a total of 15 minutes with auto steer on side roads. I have had a great experience with it on the highway but the side streets are terrifying! It immediately tried to drive into the first car that passed me, next it turned hard towards a fire hydrant and it did a rapid swerve when crossing an intersection. I could not leave it for more than 1-2 minutes without some erratic driving experience occurring. It is downright dangerous and in my opinion needs to be disabled. I could not even log all the bug reports as there were so many and I was afraid to move my hands from the wheel. I would rather them focus their energy on getting the highway right and at higher speeds. Local roads have a Looong way to go and be careful if using it.
 
I have driven about 3-4 times for a total of 15 minutes with auto steer on side roads. I have had a great experience with it on the highway but the side streets are terrifying! It immediately tried to drive into the first car that passed me, next it turned hard towards a fire hydrant and it did a rapid swerve when crossing an intersection. I could not leave it for more than 1-2 minutes without some erratic driving experience occurring. It is downright dangerous and in my opinion needs to be disabled. I could not even log all the bug reports as there were so many and I was afraid to move my hands from the wheel. I would rather them focus their energy on getting the highway right and at higher speeds. Local roads have a Looong way to go and be careful if using it.

Dont they need this exact data in order to improve it though?

It sounds like the hands should not be removed from the steering wheel anyway in this beta version, so the fact that you were afraid to is probably a good thing.
 
  • Like
Reactions: Bet TSLA
Dont they need this exact data in order to improve it though?

Half the people believe that it's gathering data all the time and AP doesn't need to be on to learn. In which case they don't need this data, they just need you driving and it watches in shadow mode and thinks "well, I would have veered left if I was turned on but the driver didn't," and learns from that.

Others believe it only effectively learns when engaged.

Let's hope the first is true, because if it's not, getting "hundreds of millions of miles" on AP2 cars is going to take years because AP2 is so restricted right now.
 
I have just the opposite experience. My AP2 car is in the shop and I got a 2015 AP1 P85D loaner. Drove it for 2 weeks for a business trip in NJ. I wanted to experience as much AP1 as possible so I enabled it every chance I got. The few most dangerous incidents were all at more than 45 degree bends. It literally going straight to the other side of the road. Tried a few more times everyday at the same spot and hope it would learn but it behaved exactly the same. It was a quiet road and hardly have any cars. Otherwise I wouldn't dare to try that again.

Are there hardware differences between the S and the X that might account for the differences?
 
Generally Autosteer will only work going straight in a clearly marked center lane.
That's pretty much AP1 behavior, it would work well in any lane but only on straight or slightly curvy lanes.

Anytime you are at a stop sign and no lead car in front, disengage AP and reengage after crossing it.

At an intersection with traffic light, disengage AP if you are the front most car.

Approaching a turn, disengage AP and take the turn yourself including the acceleration.

I guess Tesla is shooting themselves in the foot by showing off hyper capabilities of AP2 (over promising) and delivering updates so slowly (underdeliver), that as each day passes and AP2 doesn't get better, it only worries me if Tesla sold something to people that they wouldn't be able to deliver in the committed time frames, or perhaps never. I am in the software world too and I tend to agree with other folks here, even if you are on a private code branch the codebase has to be similar somewhere!

Perhaps time to ask musk himself on Twitter as why there's huge gap in the functionality.

On local roads, our AP1 car slows for turns, easily handles 90 degree bends staying perfectly in the lane.

Do you have a video of the car taking a 90 degree turn while on AP I don't think it is capable of that.
 
I rather suspect that the Tesla AP group is is high pressure place to work right now. :) And I'm quite sure that Tesla would have preferred to continue down the road further with Mobileye, had it not been that ME was trying to blackmail.... sorry... ENCOURAGE, them to sign up with ME for eternity.

I'm not surprised at how this has gone. Of course we won't know the full story until another Musk autobio in a decade or so, but I expect that we'll find out that Musk had told ME to stuff it, while the the Tesla autopilot software was virtually non-existent. And then they had to go from zero to operational in a matter of a few months. I imagine that the walls were ringing when the software team learned what happened!

The good news is that this will progressively get better, and in 3-4 months AP2 will be noticeably better than 1.0. But... I stand by my opinion that we're going to be 20 years, and a bunch of enabling infrastructure spending, before we see really usable L5 systems.
 
  • Informative
Reactions: davidc18
We're nearly at 4 months since the EAP announcement of 2-3 months with the order page telling us EAP was expected to be done in December. Starting on 12/31, they've pushed out half-baked and anecdotally dangerous updates. Musk has several times tweeted deadlines that have been missed. The Product Specialists and Owner Advisors, for let's hope in just simple ignorant exuberance, tell customers things that were/are simply false.

I got my car in December and was not familiar with this as some sort of routine Tesla way of doing things that for some reason we were supposed to grin and bear.

All this, and yet I was still totally disappointed that I didn't get an update notification last night. :eek:o_O:(
 
We're nearly at 4 months since the EAP announcement of 2-3 months with the order page telling us EAP was expected to be done in December. Starting on 12/31, they've pushed out half-baked and anecdotally dangerous updates. Musk has several times tweeted deadlines that have been missed. The Product Specialists and Owner Advisors, for let's hope in just simple ignorant exuberance, tell customers things that were/are simply false.

I got my car in December and was not familiar with this as some sort of routine Tesla way of doing things that for some reason we were supposed to grin and bear.

All this, and yet I was still totally disappointed that I didn't get an update notification last night. :eek:o_O:(

Well, initially there were many on the forum that made excuses or claimed "you just don't know how Tesla works!". But for the rest of us, the situation was revealed in the actual presentations and videos released. Some of us sort wheat from chaff for a living, and the clues that hype was the only thing being revealed were pretty clear.

But, I don't think complaints will serve much purpose at this point. That Tesla isn't currently capable of duplicating AP1 performance, let alone "enhanced" AP is now self-evident. What would you like them to do? They are working to recover as quickly as possible. Did they grossly overstate their capability? Yes. Did they hype the situation to cover up the huge setback that Mobileye's departure created? It appears so. Did they knowingly mislead customers? Obviously.

But they are doing what they should to develop their own system as quickly as possible. You can always buy a CPO AP1 car until this situation is resolved. If another maker starts shipping Waymo-equipped cars, and Tesla is still as far from "full self driving" as they appear to be now, buy a Waymo car. Until such an option is actually in the market, my suggesting is relax and enjoy what you have.