.48 feels like AP2 finally passed AP1

[Joe F]

I like jokes that are sufficiently poignant. I like a good touche of course. This one IMO missed the mark.

This one implied if only I drove more the I'd have California-level mapping or conditions for EU in my car.

Obviously a false premise. One can consider that funny perhaps if they actually don't believe in the regional difference angle. To me it is a frustrating reality that seems to make exchanging experiences here really hard, because you get sunny state people and U.S. mapped/conditions experiences that are very different from snowy RoW.

Okay, so now that the holiday spirits have let me with a hang^H^H^H headache, I will have to remind you of the maps thread. Clearly it proves that road segments contain a confidence factor. This can only be generated, and increased, by successive Tesla AP2 traversals. How else would said information populate the splines?

Given the above, yes, the more heavily Tesla populated a region is, the better mapped it will be, and the higher the confidence level will be. I made this argument many moons ago, based on common sense, but then it was very early on in AP2 having been turned on, and my less than stellar experience with the roads I traveled may have been due to AP baby steps.

So while not at the time intended to be funny, my point was and still is, remembering your own words of frequent use of an ICE rather than your MX, the solution to your experience is yours to provide.

Now, go forth and drive. Or, convince your neighbors in (insert your country here) that you can save them $1K on their Tesla purchase, and they too can help improve your Tesla experience.

And, hopefully by now the elves have dropped 50.2 in you neck of the woods.

Oh, almost forgot:

 

[croman]

No 1k discount.

 

[AnxietyRanger]

So while not at the time intended to be funny, my point was and still is, remembering your own words of frequent use of an ICE rather than your MX, the solution to your experience is yours to provide.

@Joe F I appreciate some of our recent exchanges, so worry not, the joke was not completely lost on me and no hard feelings. Hopefully not on either side.

That said, California does have A LOT of Tesla. No amount of myself driving can compensate for that difference. But I do get the idea of your joke, even if I think it missed the mark a little...

And, hopefully by now the elves have dropped 50.2 in you neck of the woods.

They have not, unfortunately.

 

[Joe F]

No 1k discount.

Ooops, thanks. Guess I need to pay closer attention to their marketing changes. Free Supercharging for life it is.

 

[Joe F]

@Joe F I appreciate some of our recent exchanges, so worry not, the joke was not completely lost on me and no hard feelings. Hopefully not on either side.

No, no hard feelings at all. After all, we're all here to share a common goal, and to help make the transition to BEV's an attainable realization.

Don't give up on Elon's elves. 50.2 is at 506 on the tracker, so it's an all out deployment. 865 is the high water make for a .42 variant, so with a WiFi connection it can't be long now. Unless you've been naughty, then all bets are off.

 

[AnxietyRanger]

Unless you've been naughty, then all bets are off.

That is always a possibility.

 

[daktari]

I am not sure that maps are utilized yet in my area, Oslo. A city of 600k, probably one of the denser populated Tesla cities of the world.
My videos have tight turns and crests, similar to @buttershrimp's latest video, but with a divider. You can see it drift left towards the fender at the end, approx 5 minutes in.
I rate .38 the highest since the following had too big troubles with slopes. 50.2 seemed very good tonight at the autobahn when lanes were clear, but was unusable (swerving terribly) when not and still slopes/undulations troubles it a lot. Too different conditions to make a definite judgement.

 

[buttershrimp]

I am not sure that maps are utilized yet in my area, Oslo. A city of 600k, probably one of the denser populated Tesla cities of the world.
My videos have tight turns and crests, similar to @buttershrimp's latest video, but with a divider. You can see it drift left towards the fender at the end, approx 5 minutes in.
I rate .38 the highest since the following had too big troubles with slopes. 50.2 seemed very good tonight at the autobahn when lanes were clear, but was unusable (swerving terribly) when not and still slopes/undulations troubles it a lot. Too different conditions to make a definite judgement.

Yes, I had some luck and then the Easter egg update dropped and another update and it’s like I went back to .32 or something . I’m sure I was just on a good road when I thought it was better, it doesn’t make sense that it would be different software on a same number update. Frustrating for sure.

 

[jimmy_d]

Ok, getting back to the topic at hand -

Drove AP2 48 from northern LA to San Diego on Xmas eve (210 to 57 to 5), which is 120 miles of crappy white on white lane markers (see attached image for example). 100% of it was on autosteer except for freeway changes. And... zero interventions. I think that might be a first for me - doing an entire non-trivial drive without any interventions required.

So these roads are kind of in the sweet spot of where I find AP2 outperforming AP1 - namely poor lane markers but otherwise tolerable roads. This particular drive was mostly straight into the sun, which is part of why the snapshot from my phone above is so crappy, but it's a fair evaluation of the majority of the drive.

I guess this is the San Diego autopilot version of a white Xmas - white lane lines on white concrete. Anyway pretty happy with that particular performance but now I've been upgraded to 50.2 - does this mean I have to start a new thread? Continuing the road trip tomorrow with a drive up to the top of Palomar mountain so I guess I'll get to check out performance on curvy roads. My hopes are not high but we will see.

 

[buttershrimp]

View attachment 268680
Ok, getting back to the topic at hand -

Drove AP2 48 from northern LA to San Diego on Xmas eve (210 to 57 to 5), which is 120 miles of crappy white on white lane markers (see attached image for example). 100% of it was on autosteer except for freeway changes. And... zero interventions. I think that might be a first for me - doing an entire non-trivial drive without any interventions required.

So these roads are kind of in the sweet spot of where I find AP2 outperforming AP1 - namely poor lane markers but otherwise tolerable roads. This particular drive was mostly straight into the sun, which is part of why the snapshot from my phone above is so crappy, but it's a fair evaluation of the majority of the drive.

I guess this is the San Diego autopilot version of a white Xmas - white lane lines on white concrete. Anyway pretty happy with that particular performance but now I've been upgraded to 50.2 - does this mean I have to start a new thread? Continuing the road trip tomorrow with a drive up to the top of Palomar mountain so I guess I'll get to check out performance on curvy roads. My hopes are not high but we will see.

This is good news!

 

[Joe F]

I'll post this here instead of a dedicated 50.2 thread. My limited experience on a highway with sweeping curves at speeds somewhat above the posted 55 MPH, was not good at all with 50.2. While I don't have a problem letting the car drift with nothing in an adjacent lane, I'n not a fan of driving in the debris that collects on outside lanes. I can't say for sure if it would have crossed the line as I nudged it back which forced a disengagement.

However, I'm really not a fan of being an inch or two from a lane marker when there's a car inches away. I had been on .44, and the jump to 50.2 feels like a major regression WRT curves at speed. I clocked major disengagements on the same road later at night, and would have thought the increased marker contrast would have fared better. Fortunately there was little to no truck traffic being Christmas...

On the plus side, straight line micro adjustments were all but gone vs. .44.

I've never gotten a camera needs calibration type delay with AP, FWIW. I know some post 12/2016 produced AP2 have.

Needs more road time to say for sure. Hopefully my bug reporting it the second time it happened, not to mention all the disengagements that hopefully triggered feedback helps. Will follow up via mail today.

 

[croman]

I'll post this here instead of a dedicated 50.2 thread. My limited experience on a highway with sweeping curves at speeds somewhat above the posted 55 MPH, was not good at all with 50.2. While I don't have a problem letting the car drift with nothing in an adjacent lane, I'n not a fan of driving in the debris that collects on outside lanes. I can't say for sure if it would have crossed the line as I nudged it back which forced a disengagement.

However, I'm really not a fan of being an inch or two from a lane marker when there's a car inches away. I had been on .44, and the jump to 50.2 feels like a major regression WRT curves at speed. I clocked major disengagements on the same road later at night, and would have thought the increased marker contrast would have fared better. Fortunately there was little to no truck traffic being Christmas...

On the plus side, straight line micro adjustments were all but gone vs. .44.

I've never gotten a camera needs calibration type delay with AP, FWIW. I know some post 12/2016 produced AP2 have.

Needs more road time to say for sure. Hopefully my bug reporting it the second time it happened, not to mention all the disengagements that hopefully triggered feedback helps. Will follow up via mail today.

Mine was better about this after a day. 50.2 still blows. Major regression. Its dangerous. I can't trust it. It was so much better after .42 and now its back to being unreliable. Curious as to @jimmy_d's impressions. Also it cannot handle any worn or faded lane lines now. It was far more robust earlier.

When lane lines are perfect it does a good job except on curves (where it disregards lane lines and lazily goes to the outside of curves for me).

 

[buttershrimp]

I'll post this here instead of a dedicated 50.2 thread. My limited experience on a highway with sweeping curves at speeds somewhat above the posted 55 MPH, was not good at all with 50.2. While I don't have a problem letting the car drift with nothing in an adjacent lane, I'n not a fan of driving in the debris that collects on outside lanes. I can't say for sure if it would have crossed the line as I nudged it back which forced a disengagement.

However, I'm really not a fan of being an inch or two from a lane marker when there's a car inches away. I had been on .44, and the jump to 50.2 feels like a major regression WRT curves at speed. I clocked major disengagements on the same road later at night, and would have thought the increased marker contrast would have fared better. Fortunately there was little to no truck traffic being Christmas...

On the plus side, straight line micro adjustments were all but gone vs. .44.

I've never gotten a camera needs calibration type delay with AP, FWIW. I know some post 12/2016 produced AP2 have.

Needs more road time to say for sure. Hopefully my bug reporting it the second time it happened, not to mention all the disengagements that hopefully triggered feedback helps. Will follow up via mail today.

@croman I’m experiencing the exact same thing... major regression in several aspects of autosteer. On the flip side it’s getting much better the more I drive? Could we be looking at a more robust calibration process? For example, first day, I could not navigate previously navigable roads and second day I can... the swerving on hills and crests is still very poor . I hate saying this is a regression but I’d have no street cred if I don’t point out this firmware has thus far been not safe or reliable. I hope Tesla will instruct folks if there is a calibration period at new firmware. I’m about to do another drive and I’ll report back.

 

[AnxietyRanger]

I respect your candor @buttershrimp.

On that note, got .50.2, but due to Christmas haven't been going anywhere yet. I will report back once I do.

 

[croman]

@buttershrimp -- I have noticed it is doing better but its not anywhere near as good as .42 (driven over 400 miles these past 3 days despite snow and extreme cold (it was -8F last night!!). My lifetime wh/mi is taking a beating but its not because AP fails to understand its too cold get regen for short distances.

Anyways, I took another trip to the loop (downtown Chicago) and 90/94 (the largest/most heavily trafficked of 3 highways to get to Chicago) was still a hot mess. All the curves were better but it was still dangerous and loose and would've crossed lane lines twice. It was worse than 17.17.4. We're talking 17.11.3 bad (in regards to curves).

17.17.4 was herky and jerky (8 bit curves) but it kept the car safe. 17.42 and on until now was the greatest combo of competence and smoothness. Why is it a tradeoff?

 

[AnxietyRanger]

@verygreen has the NN changed between recent versions?

 

[verygreen]

@verygreen has the NN changed between recent versions?

In 17.46 or 17.48 they split out pillar NNs from repeater NNs, the network is about the same with just tweaked weights.
17.50 NN is same as 17.48

 

[BLKTSLA]

My year-end observation....

When Tesla releases a significant AP update they will state it in the release notes, e.g. major update, or minor updates to AP. If its not in the release notes expect general updates to progress and regress with each version. I can't say I understand why it would regress, but that has been the case and seemingly consistent based on discussions here. That said, going into 2018 I expect there to be one of the major AP updates at in the beginning of the year most likely to coincide with the update for Tesla Maps. I predict that following that update we will see another major update towards AP+ and FSD before the end of the 1st quarter.

 

[AnxietyRanger]

@buttershrimp OK, first .50.2. motorway experiences:

- Overall working OK
- Lane positioning seems to have improved at least on one stretch
- Steering wheel movements seem less relaxed, a bit more erratic or forceful in .50.2. than they were in .46.8.

 

[arcus]

@buttershrimp OK, first .50.2. motorway experiences:

- Overall working OK
- Lane positioning seems to have improved at least on one stretch
- Steering wheel movements seem less relaxed, a bit more erratic or forceful in .50.2. than they were in .46.8.

Same experiences here. I would add that auto lane change seem to be more aggressive than in 2017.46 (at least left-to-right change).