Why are TACC, AP (and ?FSD) so bad?

[beatle]

I wish Tesla continued to use Mobile Eye's system for boring freeway drives and continued their work on FSD for local streets.

Even if they stuck with their own hardware and just focused on highway behavior until they moved up the autonomy stack on divided highways, I think we'd be in better shape. Programming for city navigation is a waste of time until they've got the "easy" environment done.

 

[mspohr]

AP1 owner here (2015 Model S) and can attest to how good it works on the freeway. It really is stress reducing and way better than my mother's 2021 Subaru Forester Eyesight. Her car ping pongs even with lane centering, constantly nags and beeps at you, and completely gives up steering on any slight bend on a back road. At least AP1 tries real hard to maintain the centering and does so quietly.

I wish Tesla continued to use Mobile Eye's system for boring freeway drives and continued their work on FSD for local streets.

I had a 2015 Model S. The AP and TACC were much better.
None of these problems.

 

[AndrewZ]

Mobileye had broken their partnership with Tesla back in the day after the earliest Autopilot fatalities and Tesla was forced to try rebuilding what was lost, otherwise it probably would still be in use. IIRC Elon was quoted back then saying that it was very easy to build back, there might still be tweets saying that.

Mostly happened behind closed doors but my digging suggested Mobileye was at odds with pushing the implementation on public roads and Tesla's risk tolerance

 

[mspohr]

Mobileye had broken their partnership with Tesla back in the day after the earliest Autopilot fatalities and Tesla was forced to try rebuilding what was lost, otherwise it probably would still be in use. IIRC Elon was quoted back then saying that it was very easy to build back, there might still be tweets saying that.

Mostly happened behind closed doors but my digging suggested Mobileye was at odds with pushing the implementation on public roads and Tesla's risk tolerance

This is generally in line with Tesla's policy of vertical integration. They do as much as possible in house to avoid relying on outside contractors.
They probably didn't realize how much work it would take to bring their own systems up to par with Mobileye.

 

[E90alex]

Well then TACC is a TERRIBLE L2 driver assist. My mistake for assuming it was as good at NOT slamming on the brakes as our Volt, Caddy and Mustang Adaptive cruise control on regular roads. NEVER had phantom braking before on those and didn't even know Phantom Braking was a thing until got a Tesla. And it sucks.

To answer the question, Tesla can't do as good a job as Mobile Eye at controlling the speed of a L2 car. Before Tesla got into a pizzing match and wanted to over sell the Mobile Eye based AP back in 2016, Tesla used Mobile Eye and it was good. Then TechNOking oversold it, Mobile Eye pulled themselves from Tesla, and Tesla hasn't been able to do as good a job since on their own. Atleast that's my understanding from descriptions of MS owners than had the old, good Mobile Eye system.

RSEV on YouTube did a test of auto park systems a while ago with various EVs. Other cars using MobilEye and old Model S with MobilEye were flawless and newer Teslas were a joke.

 

[2101Guy]

Mobileye had broken their partnership with Tesla back in the day after the earliest Autopilot fatalities

Wait..I thought Elon said there has never ever ever been a fatality associated with Tesla's autonomous driving capabilities? Not just FSD, but ANY of the Tesla variants of autonomous driving capabilities..

 

[Twiglett]

I had a 2015 Model S. The AP and TACC were much better.
None of these problems.

funnily enough I'm probably heading in that direction.
I need a second car with a bit more space but don't want to go Y or X, so a used S with AP1 is what I'm shopping for in the next few months
I've got FSD beta on my Model 3 and it is sometimes spectacular, but other times is appallingly bad.
I'd much rather have what AP used to be, it couldn't do everything, but what it could do it did pretty well.

 

[mspohr]

Tesla recalls 362,000 vehicles over self-driving software flaws that risk crashes

Regulators say driver assistance system does not adequately adhere to traffic safety laws and can cause crashes

www.theguardian.com

NHTSA said “the system may respond insufficiently to changes in posted speed limits or not adequately account for the driver’s adjustment of the vehicle’s speed to exceed posted speed limits.”Self-Driving (FSD) Beta software after regulators said on Thursday the driver assistance system did not adequately adhere to traffic safety laws and could cause crashes.Tesla said in “certain rare circumstances … the feature could potentially infringe upon local traffic laws or customs while executing certain driving maneuvers”.

 

[DrChaos]

Two lane highway (most recent CA 190). This is a winding road with a speed limit of 55 for the curvy parts. Gentle curves no problem. Sharper curves (those with the big yellow speed advisory signs) it slows down but not enough to successfully navigate the curve and ends up going out of the lane into the oncoming lane or off the edge of the road.
The fact that it slows down tells me that there must be something in its little brain that knows it's a curve and that it should slow down but it's not able to calibrate the speed to the curve.

A two lane highway (vs a divided CA freeway where there are standards on how much a road can curve or not) isn't so much in the AP training set.

It tries but that is more of the domain of FSD which I know can slow down significantly (its usual problem is too much). CA 190 is rural going into Death Valley, less frequently cars ahead to follow. I'm driving in San Diego where there are always other cars on the freeway---AP/FSD locks in better when it sees tail of other cars.

Machine learning methods are sensitive to the distribution of their training set. It's unquestionable that most of the CA Tesla training set for AP will be Bay Area, LA, and San Diego freeways in the city and suburbs, as that's where most of the people are and where they drive. There's (wag) 10000 Tesla drives in LA freeway for every one on CA190.

Personally if I were using the AP out of domain like that, I would keep my finger on the right wheel button with speed and adjust it to drive like person by planning ahead. The AP cameras are not as good as human vision, its "think ahead time" is less than a human's.

If you had Ford or GM's BrownCruise it would have given up completely there and made you drive 100% manual.

 

[dgatwood]

This would probably be worth testing after FSD Beta v11 rolls out fully and gets merged into the mainstream autopilot stack. Right now, TACC and basic autopilot are using the highway stack. Once FSDb takes over highway autopilot duties on FSD beta, presumably there will be some magic cutover version where that also becomes true for people who just have TACC and basic autopilot (i.e. exits and turns will be turned off, but the core path planning functionality will use the newer stack). At that point, I suspect all of the problems you describe will go away, but we'll see.

 

[jmcdono362]

Mobileye had broken their partnership with Tesla back in the day after the earliest Autopilot fatalities and Tesla was forced to try rebuilding what was lost, otherwise it probably would still be in use. IIRC Elon was quoted back then saying that it was very easy to build back, there might still be tweets saying that.

Mostly happened behind closed doors but my digging suggested Mobileye was at odds with pushing the implementation on public roads and Tesla's risk tolerance

I could see that being the case. AP1 does try hard on city streets and pretty darn good at it considering nothing else on the market at the time came close.

 

[jmcdono362]

RSEV on YouTube did a test of auto park systems a while ago with various EVs. Other cars using MobilEye and old Model S with MobilEye were flawless and newer Teslas were a joke.

AP1 auto park pretty much just goes forward and backwards, but does it well. Except for my driveway which is super steep, which it refuses to do. It even says so on the app why it stopped.

 

[jmcdono362]

funnily enough I'm probably heading in that direction.
I need a second car with a bit more space but don't want to go Y or X, so a used S with AP1 is what I'm shopping for in the next few months
I've got FSD beta on my Model 3 and it is sometimes spectacular, but other times is appallingly bad.
I'd much rather have what AP used to be, it couldn't do everything, but what it could do it did pretty well.

Keep an eye out for a 2015 with the 90 kwh pack. Kind of a rarity but it's the sweet spot for reliability, AP1, and free supercharging/connectivity. I've had mine for 4 years and driven 50K miles It's been nearly flawless aside from 2 door handles, tires and suspension maintenance for creaking noises.

Did a 1500 mile trip from Boston to FL this winter and it was a pure joy when 90% of the drive is on autopilot. Still get 200 miles with warm weather and low winds.

 

[zoomer0056]

TACC and AP

Just a nitpick here: as per the Tesla Model S manual Autopilot is the name of a group of features of which one is TACC. It's incorrect to say "TACC and AP" as TACC is one feature of AP.

 

[PVia]

It's useless to teach anyone here the names of the features or the proper usage of them...
Also, everyone on the forum knows exactly what's going on in the Autopilot division, how they're being lazy and shuffling their feet on V11...
Many of them are upset that Tesla hasn't responded with a fix to their issue when they pushed their little camera button...
Every day more complaints about anything and everything pile up on these boards with their whiny voices, snivelly noses and first world entitled problems...it's getting really old, bitches.

 

[FirstInTown]

It's useless to teach anyone here the names of the features or the proper usage of them...
Also, everyone on the forum knows exactly what's going on in the Autopilot division, how they're being lazy and shuffling their feet on V11...
Many of them are upset that Tesla hasn't responded with a fix to their issue when they pushed their little camera button...
Every day more complaints about anything and everything pile up on these boards with their whiny voices, snivelly noses and first world entitled problems...it's getting really old, bitches.

No, not unreasonable to have this P. O. S. Phantom Braking fixed. It had been a problem for YEARS (since 2016 spilt with mobile eye) It isn't whining when my expensive MX slams in the brakes unexpectedly so hard my golden retriever falls forward off the back seat. That is just terrible and none of our 3 other cars with ACC cruise do this. The Tesla's is just currently garbage.

 

[mspohr]

Just a nitpick here: as per the Tesla Model S manual Autopilot is the name of a group of features of which one is TACC. It's incorrect to say "TACC and AP" as TACC is one feature of AP.View attachment 908614

My 2022 S LR only has the first two items on this list. (And they don't work)

 

[zoomer0056]

My 2022 S LR only has the first two items on this list. (And they don't work)

My Feb 2022 MSLR has your same set up. I would say they work badly. After a couple of phantom braking I decided to not use TACC and I just don't trust autosteer. It's way too nerve racking to drive on edge expecting it to mess up. I enjoy driving my S myself. It makes no sense to me that this beta software is allowed in the general public. When I see another Tesla near me I give it lots of room.

 

[OxBrew]

If AP is not meant for 2 lane roads, why does it activate on 2 lane roads? I should be fenced, if it's unsafe.

Seriously, I really don't understand that.

 

[PVia]

Don't know how to answer that, except RTFM.
Also, if a wheel weight is not supposed to be used for hands-off driving, why do people still use them?
If you're required to keep your hands on the wheel, why do people post videos of themselves sleeping or having sex while driving?
People do so many stupid things that make it difficult for the rest of us, regardless of guidelines, manuals and common sense.