FSD is not good enough for me. Sorry

[CG Cowboy]

It's fascinating how your post (and others) trigger a visceral response. I couldn't agree with you more.

First of all, it's completely unnecessary to attack someone for their opinion. It wasn't negatively targeting anyone in these forums. Agree or disagree. Share counter opinions. Sharing opinions , experiences, and ideas is why we're here. But share in a way that doesn't shut people down.

Secondly, I think the way you worded the heading is spot on. "Not Good Enough" to me means it's not the standard I would have expected. Is it perfect? Of course not. No one expects that. But this is not the first iteration of Tesla's FSD. For the CT, yes. But the issues being reported don't appear to be issues that would be caused by the differences in the CT relative to the other Tesla models using FSD. I'm having a blast trying it out and it's doing well in plenty of scenarios. But where it's not performing well, I would have expected it to perform better. Not good enough for now. But I'm looking forward to the improvements.

Sponsored

 

[CyberGus]

Each time I intervene, I'm prompted to record a short voice note and explain why I disengaged. I try to make the note each time, with the expectation that the video/CT data and my commentary makes it back to the engineering team and helps contribute to future improved releases. Are you other FSD "supervisors" doing this as well? Is this normal practice for all models, or just CT?

I don't bother explaining that I had a sudden urge for drive-thru and veered off course lol. A few times I disengaged because I was still learning how to trust.

But when I'm forced to take over? Yeah, I leave a message. Hopefully not too snarky

 

[Paul Marsh]

good thing you didn't use it a couple years ago lol

It's about 100 times better then it used to be

Pretty sure the Robotaxi will be geofenced and have more sensors than the current cars.

Warning: Please be extremely careful in school zones. FSD does not recognize flashing signals and mandatory 20 MPH school zones.

 

[Jhodgesatmb]

I understand it's the best thing out there for the moment in terms of being close to autonomous but it certainly isn't good enough yet to drive itself without interventions in urban areas. If Robotaxis is the same as FSD 12.5.5, it would be highly risky I think. I love Tesla so I really want it to be good but it cuts corners too sharply and also misses things that it cannot classify such as plastic roadsigns that stick up in the middle of the street (it does see cones and other objects though). As an example of a dangerous scenario, mine made a turn and instead of going on the road, it went onto a local light rail track and followed that path instead which could be disastrous if one happened to have been oncoming.

I agree that it isn't very good, at least not any better than before, but the important thing is that there is now a version on the Cybertruck and, from here on, it will get incrementally better.

 

[Crissa]

Contrary to popular belief you can’t learn everything from the internet. I don’t need to write “citations.”

Look, you clearly didn't learn that FSD wasn't above Level 2 autonomy from an authoritative source not-the-internet.

I, on the other hand, see the point of the internet is to learn.

FSD will disengage right before impact blaming you for this accident saying see FSD wasn't even on when you crashed.

This is untrue. Any accident within five seconds of disengaging is considered an autopilot or fsd accident. At freeway speed, five seconds you've gone a tenth of a mile, and enough time to have come to a full stop. And for it to disengage and kick you out it will have given you multiple warnings to take over.

There are plenty of videos of FSD hitting cars, medians, signs, curbs, going the wrong way, etc.

There's not, in fact, alot of video of that.

...and no, reddit photos are not evidence. Nor are Teslacam videos, since they do not show which automation it's in nor the error messages. Yeah, that that's a high bar!

But more than half of the incidents independently reported to NHTSA turned out to be 'overrode Autopilot with pedal' or 'weren't in that level automation at all'.

-Crissa

 

[Sirkevin]

It does not recognize stop signs for entry gates. It went right through the stop, sign and almost hit the gate.

 

[BengalBoy]

You wrote probably the most helpful and meaningful FSD experience I've read thus far.
xAI Teslas AI version has the ability to learn. I've used FSD once for only a few minutes leaving my neighborhood to run an errand. I took over after a few blocks because of an unusual situation that I wasn't going to wait to see if FSD was going to figure it out. Residential construction traffic, gardener trucks and other oddities not usually there.
When I used enhanced autopilot in my 2018 M3 it worked pretty well (only rated for hwy driving) it was close to perfect. Especially in stop and go traffic of Los Angeles.
Reading the areas you and others wrote is helpful to know what to watch for when using FSD. This is one of the real values of this forum. Now I know to beware of sharp left turns etc. Let's build on this for all. I'm very excited to help test and train FSD going forward.

Thank you. I'm happy to hear it. This morning I left my driveway on a 60 mile trip through city 4 lane traffic, turning into a county two lane highway. A different route that i had not yet driven. Coming around a blind curve leaving a school zone FSD began to speed up though a traffic light was red. I had to intervene and brake. For the 1st time a message appeared asking me to leave a voice message and send it to Tesla regarding why I disengaged FSD. About 5 miles later it happened again on the freeway as i topped a hill, the traffic was backed up from the red light and FSD was still making 70 mph with no slowing. I hit the brakes again and disengaged FSD. Another prompt hit the screen asking me to leave a voice message regarding why I disengaged FSD and to send it to Tesla. So I did. The remaining 50 miles to Greensboro, Alabama was 4 lane freeway and county two lane highway traffic through small towns, braking for logging trucks and farm tractors ? 007 completed the trip flawlessly to my destination. It seems to need a direct and unobstructed view of traffic controls and quickly coming up on an intersection with traffic lights that it hasn't had some time to see required intervention. However, Tesla is now asking the driver to explain why they disengaged FSD. It is only going to improve as we use it and help Tesla to understand our concerns. ?

 

[dalton108]

It is not as plainly obvious as you suggest. The geofencing is counteracted by much safer autonomy, hence the SAE classification. A passenger doesn't have to look at the road or hold the wheel in a Waymo. In this context, FSD is not "viable" at all, by comparison. At least not yet.

It is not a party trick when it reliably and safely gets you around your city, as it does for me.

The last road trip I did before I took my delivery of my cyber truck was in my 2023 Model S Plaid on FSD 12.5.1. The car drove us round-trip from my home to our hotel outside of Dana point more than 500 miles round-trip with me doing about 5% of the driving. The rest of the time since this was the first vision based monitoring system I was able to literally sit in the car with my arms folded.

I’m not sure how you measure value and I hear you people all the time talking about how it doesn’t work it’s dangerous etc. but being chauffeured 500 miles anytime I want to is simply invaluable to me. And it’s not a future that I’m waiting for it’s a reality that I’ve been experiencing since July of this year with the release of vision based FSD monitoring. Now I have that on my Cybertruck. My wife has it on her new model X. My dad has it on his 21 model Y performance (he’s 75 years old and couldn’t be happier with it). We live in alternate realities.

In my world, I can put my hands on four vehicles in my driveway that can with very little supervision take any driver with eyes and the ability to look straight anywhere they want to go with nary an intervention.

in the world that some of you seem to live in the same product that I can show you a picture of a 75-year-old man who uses it with aplomb - does not work, is scary and dangerous and a waste of your money/fraud.

??‍

I guess everybody in my family is just lucky, smarter and/or more capable. For many years the software has been less than perfect it’s not perfect now but my God it’s so good as to be, at times,: indistinguishable from a human driver so I don’t get it.

But I, for one, I am perfectly OK with other people depriving themselves of these and any other advantages the world has to offer them out of fear, ignorance or confusion.

Your loss is my advantage.

 

[dalton108]

It does not recognize stop signs for entry gates. It went right through the stop, sign and almost hit the gate.

Really good thing that there was a driver responsible for operating the vehicle present then, huh? ?

 

[jxmpd]

The last road trip I did before I took my delivery of my cyber truck was in my 2023 Model S Plaid on FSD 12.5.1. The car drove us round-trip from my home to our hotel outside of Dana point more than 500 miles round-trip with me doing about 5% of the driving. The rest of the time since this was the first vision based monitoring system I was able to literally sit in the car with my arms folded.

I’m not sure how you measure value and I hear you people all the time talking about how it doesn’t work it’s dangerous etc. but being chauffeured 500 miles anytime I want to is simply invaluable to me. And it’s not a future that I’m waiting for it’s a reality that I’ve been experiencing since July of this year with the release of vision based FSD monitoring. Now I have that on my Cybertruck. My wife has it on her new model X. My dad has it on his 21 model Y performance (he’s 75 years old and couldn’t be happier with it). We live in alternate realities.

In my world, I can put my hands on four vehicles in my driveway that can with very little supervision take any driver with eyes and the ability to look straight anywhere they want to go with nary an intervention.

in the world that some of you seem to live in the same product that I can show you a picture of a 75-year-old man who uses it with aplomb - does not work, is scary and dangerous and a waste of your money/fraud.

??‍

I guess everybody in my family is just lucky, smarter and/or more capable. For many years the software has been less than perfect it’s not perfect now but my God it’s so good as to be, at times,: indistinguishable from a human driver so I don’t get it.

But I, for one, I am perfectly OK with other people depriving themselves of these and any other advantages the world has to offer them out of fear, ignorance or confusion.

Your loss is my advantage.

With all due respect, I'm not sure why this was a reply to me. I think FSD is cool and has value.

I was only pushing back against this part of the OP:

"I understand it's the best thing out there for the moment in terms of being close to autonomous"

 

[rlhamil]

Imo Robotaxi is years away and is just going to be some BS reveal like the Roadster and humanoid robot. There's no way Tesla has a product capable of working as well as Waymo right now, just another stunt by Elon to keep the stock moving and interested in Tesla

Keeping the stock moving is how they can afford to eventually (even if years late) deliver. They're not JUST making stuff up, they're making it happen.

So yes, you're being manipulated, but not just for a quick buck; for actual results, one day. So work real hard on staying alive until that day arrives.

 

[rlhamil]

oh for sure, I completely understand the software development lifecycle and the several nuances and dynamics regarding FSD and how it will mature over the next few years.

Basically what I am looking for is a real, solid video example of a collision that is actually caused by FSD where the person behind the wheel didn't have enough time to overtake. So far the only examples I have seen are situations in which a human nor machine would be able to overcome the law of physics and a crash was inevitable anyway.

So far, I haven't seen such an example. Granted, I'm almost positive there are situations out there that absolutely do fit that criteria.

There probably are, but not a lot.

A reasonably skillful human can look farther ahead and provisionally plan moves farther ahead (ready to adjust if something changes). I don't think the AI does that. The nav "plans", but the driving just deals with situations, and the farther out is included in the situation, the less often it will have already been encountered.

 

[rlhamil]

With all due respect, I'm not sure why this was a reply to me. I think FSD is cool and has value.

I was only pushing back against this part of the OP:

"I understand it's the best thing out there for the moment in terms of being close to autonomous"

Most of those that are clearly better are geofenced or have remote assistance or some other special case or restriction. One or two that are equally general MIGHT be better (Mercedes?), but that's about it.

 

[jxmpd]

Most of those that are clearly better are geofenced or have remote assistance or some other special case or restriction. One or two that are equally general MIGHT be better (Mercedes?), but that's about it.

Yes it's geofenced, but it is doing things no one else is doing commercially. Writing off Waymo because it's geofenced (so far) is like ignoring FSD because it's not SAE level 3 or 4 (so far).

 

[rlhamil]

Yes it's geofenced, but it is doing things no one else is doing commercially. Writing off Waymo because it's geofenced (so far) is like ignoring FSD because it's not SAE level 3 or 4 (so far).

I'm not writing it off, I think it's comparing apples and oranges and trying to claim one is the better fruit.

Sponsored