Navigate on Autopilot is Useless (2018.42.3)

[DarkMatter]

I remember those stories. I always wondered how people hit another car while paying attention to the road, and figured they had ceded control to the computer. That's certainly braver than me! I find AP relaxing but only because it means I can look further down the road and spend less time micro-managing my speed.

 

[drtimhill]

Target your searches from 2017-2019. There were three or four posts made about it either here or on reddit, all of them told the same exact story. Overpass that previously experienced phantom braking, no longer experienced phantom braking, didn't slow for traffic stopped under the bridge and rear ended another car.

hmm well if you make the claim its for you to provide the citations, not for me to hunt for them.

 

[DrDabbles]

No thanks. You're free to look for yourself, but I've provided more than enough evidence that you folks have intentionally pretended didn't exist.

 

[drtimhill]

No thanks. You're free to look for yourself, but I've provided more than enough evidence that you folks have intentionally pretended didn't exist.

ohhh no, that's not the way it works. There are many posts in this thread about how phantom braking has the potential to cause accidents. You then claimed that it has actually done so. That's a major, and serious, claim, and needs to be backed-up by clearly cited evidence. And yes, if you make that claim it's for you to substantiate it. That's nothing to do with "pretending it doesnt exist" .. if you can't provide reliable citations, then the one who is pretending here is you. Saying "go look it up on Google" isnt a citation, its a cop-out.

 

[DrDabbles]

You then claimed that it has actually done so.

You should re-read my comment, I guess. Because my claim is that Tesla's naive attempt at preventing phantom braking has caused collisions. We already know Phantom braking has caused collisions, that's not even being discussed here.

if you can't provide reliable citations

Yeah, except I know how you all play the game. Firs the citation is given, then you say it doesn't show what it obviously does, then you claim that the person experiencing the problem is FUDster, then you claim I'm a short for posting it, and before you know it we have a 56 page thread about a demonstrably useless feature with TONS of bugs where you're now questioning a totally different feature and how prevalent it is.

Spend your time pushing Tesla to do better.

 

[drtimhill]

Yeah, except I know how you all play the game.

I asked for citations .. three times. You avoided providing them .. three times. I'm not the one playing games.

 

[wk057]

Getting a bit off-topic here it seems. Perhaps take that discussion private?

Also, this isn't the way AP1 worked anyway as I noted in a post above:

That's actually not exactly how that worked.

The "whitelist" would store a good bit of information, including the distance and angles of the false-triggering item, aggregated over multiple vehicles. If this is an overpass, this would not match the distance and angles of a vehicle or other obstacle and thus would not cause an issue for actual safety features.

 

[cinnamon_owl]

Did it successfully take the 405S to I5S ramp? That's my biggest gripe as it hasn't changed recently. With mine it gets in the right lane for I5N, and not the lane I need to be in for I5S.

The only thing I could see being wrong is my maps version, but the fact that it lets me use NoA means it should be the latest version. I really wish Tesla would show the maps version.

There seems to be something wrong with the map themselves.

Another example would be I5N through Seattle, and avoiding the Seneca St exit.

In the Nav is said Seneca St exit, and then I5N.

Shouldn't it be stay on I5N? Why it wanting me to take the Seneca St. exit? What the nav said made no sense.

I'll have to double check this, but I recall NoA trying to get me into the Seneca St lane, but I was able to ignore it without it doing anything stupid (like braking). After I ignored it then I got the Navigation prompt that said to stay in the right two lanes (which was correct). Once I got to the tunnel then NoA said it was unavailable in the tunnel and turned itself off, and then back on once it got out of the tunnel.

After it got out of the tunnel it tried to get me out of the second to most left lane. There wasn't anything on the Maps to indicate why. I ignored it and eventually it went away. It's like it was confused about which lane I was in.

There is something wrong with how directions are being generated by the Nav. This is on a Model 3 (which should have the same maps as the Model S/X).

And SADLY, 4 years later it is STILL doing this Seneca St exit thing...along with all the other issues in the Seattle area where it thinks there are fewer lanes on the freeways than exist. This is evident when it changes from lane 2 (passing lane, not HOV) to lane 3 stating "changing lanes to follow route" then while in lane 3 changes to lane 4 stating "changing out of passing lane". (HOV = Lane 1, then lane 2 to its right, etc). I've seen this in various sections only on I5 S between Seattle and Tacoma, and on Hwy 167 between Renton and Puyallup. I basically have to just keep Nav on Autopilot off for these trips.

 

[Padelford]

Also, NoA (2022.24.8) still objects to driving NEXT to an HOV lane unless HOV lane use is enabled. I’ve complained about this for more than a year without any fix.

 

[AWDtsla]

I doubt there has been any NoA updates in a few years. I would not expect any until single stack FSD is released.

 

[cinnamon_owl]

I was thinking the same regarding the NoA updates, which I can understand at this point. So I definitely expect this to be fixed once the stack is merged.

 

[cinnamon_owl]

Also, NoA (2022.24.8) still objects to driving NEXT to an HOV lane unless HOV lane use is enabled. I’ve complained about this for more than a year without any fix.

It is really odd because I notice sections where it is perfectly fine and then now can predict exactly when it thinks it can’t be in that lane next to HOV, when I have HOV use disabled. Here’s hoping for it being fixed once merged stack is complete.

 

[AWDtsla]

I was thinking the same regarding the NoA updates, which I can understand at this point. So I definitely expect this to be fixed once the stack is merged.

That said I think something is wrong over there where everything seems to be single threaded through Elon. Old problems just fester until it's Elon's problem. If you tweet or youtube you can make it Elon's problem, if you're lucky. Took something on the order of 4 years to get the backup camera gamma on MCU2 fixed. Like, seriously? Probably a single byte fix.

 

[Padelford]

It is really odd because I notice sections where it is perfectly fine and then now can predict exactly when it thinks it can’t be in that lane next to HOV, when I have HOV use disabled. Here’s hoping for it being fixed once merged stack is complete.

Around the metro Seattle area, the bug is universally encountered. Very frustrating that I've been complaining about this by phone for more than a year, promised further contacts from developers that repeatedly never happened. Tesla seems to not give a damn about this.

 

[DrDabbles]

I doubt there has been any NoA updates in a few years. I would not expect any until single stack FSD is released.

As someone with FSD, this is going to be a real long time and when it finally gets released, I expect a substantial number of regressions. FSD is so much worse behaved than AP that I'm astonished they've released it to the public at all.

 

[Skotty]

One thing I wonder about -- there are regional differences in road markings, road materials, etc. With so many Tesla vehicles in California, and @wk057 having routes in what I would guess is a region with far less Tesla traffic, perhaps autopilot has ended up evolving to work best for California road conditions and is having more difficulty elsewhere.

 

[GWord]

As someone with FSD, this is going to be a real long time and when it finally gets released, I expect a substantial number of regressions. FSD is so much worse behaved than AP that I'm astonished they've released it to the public at all.

I recall a similar situation with autopilot hardware 2. It was a considerable regression from the Mobileye package for at least a year. Eventually, though, it did gain momentum and left the previous version in the dust.

 

[clydeiii]

Plus city driving is at least an order of magnitude more complicated and complex than highway driving, and therefore we should assume FSD will perform worse than NoA...the fact that city streets driving works at all is astonishing.

 

[drtimhill]

As someone with FSD, this is going to be a real long time and when it finally gets released, I expect a substantial number of regressions. FSD is so much worse behaved than AP that I'm astonished they've released it to the public at all.

Ever driven FSD on a long quiet road with no intersections? That's FSD on freeways, more or less. Doesnt seem any worse than AP to me, and in fact its much better at lane changes than AP, even now.

 

[DrDabbles]

Ever driven FSD on a long quiet road with no intersections?

No, because I haven't had a single FSD drive that didn't have phantom braking, intersection braking, or need me to completely take over.

That's FSD on freeways, more or less

It's not, though, because freeway speeds are vastly higher and FSD's ability to react to distant objects is not good at all.

Doesnt seem any worse than AP to me

I am absolutely giddy with anticipation for more normal consumers to get their hands on FSD. It's going to be complete chaos, but the reality will finally dawn on normal users. Not much you can do about True Believers, but I'm not particularly concerned with them anymore.