Cybertruck on full self-driving (FSD) 12.5.5.3 drives us through fog, ZERO safety interventions!

[pilotbeagle]

My 12.5.5.3 is the same crapiness. Does pretty good on the highway but:
- still turns out of my neighborhood and goes in the center opposite way turn lane
- still is pitiful at stop signs
- still turns into the center lane (again an opposite turn lane on a 5 lane road)
- still can’t plan for upcoming turns. I’m in the right lane for an upcoming right hand turn. For some reason it goes into the center lane and then signals immediately to return into the right lane ?‍
Maybe they give a dumb version for us Colorado drivers ?‍

Sponsored

 

[btcrealm]

Looks like you're a huge fan of using it for a tax write off. Maybe you should stop badmouthing an advertising money maker!

LoL, not sure how thats valid. We are talking about FSD and not the cybertruck. And no, there is no tax write off as bitcoin is an opensource technology and not a company. I have it on there to share the concept with people.

 

[Darthamerica]

My 12.5.5.3 is the same crapiness. Does pretty good on the highway but:
- still turns out of my neighborhood and goes in the center opposite way turn lane
- still is pitiful at stop signs
- still turns into the center lane (again an opposite turn lane on a 5 lane road)
- still can’t plan for upcoming turns. I’m in the right lane for an upcoming right hand turn. For some reason it goes into the center lane and then signals immediately to return into the right lane ?‍
Maybe they give a dumb version for us Colorado drivers ?‍

It’s great on most open highways and freeways. Performance degrades once you need to take an exit, in heavy traffic, on city streets and forget about parking lots! There will be many manual interventions. Nothing too dangerous, but it’s not a stress free experience yet.

The good news is that very few of the issues seem insurmountable and given another year of incremental improvements, I can see it living up to its promise by Q3/Q4 2025.

 

[ehuna]

It’s great on most open highways and freeways. Performance degrades once you need to take an exit, in heavy traffic, on city streets and forget about parking lots! There will be many manual interventions. Nothing too dangerous, but it’s not a stress free experience yet.

The good news is that very few of the issues seem insurmountable and given another year of incremental improvements, I can see it living up to its promise by Q3/Q4 2025.

I actually see it working great already on freeways, city streets, and parking lots - in both California and Oregon.

Take a look at the video below for example where Cybertruck on full self-driving (FSD) handles 4- way stops, a stopped garbage truck, and more.

It’s hard to predict an exact date, since this has never been done, but Elon Musk and the Tesla AI team has the data - and they say we’ll have unsupervised FSD in California and Texas by July 2025.

We’ll then be able to watch a movie, play video games, or take a nap while Tesla cars drive us around - it will save thousands of lives! ??

My experience from FSD version 11 to version 12 is that it got way better with the new AI models, much more human.

I’m waiting for FSD version 13 which is the result of using hundreds of thousands of the new Nvidia GPUs that Tesla bought to train the new AI neural networks - it will get even better, exciting times!

Tesla Full Self-Driving (FSD) 12.5.5 on Cybertruck HW4 City Streets - ZERO safety interventions

 

[Darthamerica]

I actually see it working great already on freeways, city streets, and parking lots - in both California and Oregon.

Take a look at the video below for example where Cybertruck on full self-driving (FSD) handles 4- way stops, a stopped garbage truck, and more.

It’s hard to predict an exact date, since this has never been done, but Elon Musk and the Tesla AI team has the data - and they say we’ll have unsupervised FSD in California and Texas by July 2025.

We’ll then be able to watch a movie, play video games, or take a nap while Tesla cars drive us around - it will save thousands of lives! ??

My experience from FSD version 11 to version 12 is that it got way better with the new AI models, much more human.

I’m waiting for FSD version 13 which is the result of using hundreds of thousands of the new Nvidia GPUs that Tesla bought to train the new AI neural networks - it will get even better, exciting times!

Tesla Full Self-Driving (FSD) 12.5.5 on Cybertruck HW4 City Streets - ZERO safety interventions

I use it everyday. I'm about to use it after I write this post. It works "ok". OK= good enough to be FSD Supervised. But it is nowhere near unsupervised FSD, watching movies or anything like that. I do understand how the technology works and I think by the later part of next year, some form on Level 3 capability in a geofenced area is possible. Maybe on some LA Freeways for example. But I think it will take a bit longer before it is capable of broader unsupervised use cases. I would say ~2027 is more realistic for that.

 

[Jhodgesatmb]

My 12.5.5.3 is the same crapiness. Does pretty good on the highway but:
- still turns out of my neighborhood and goes in the center opposite way turn lane
- still is pitiful at stop signs
- still turns into the center lane (again an opposite turn lane on a 5 lane road)
- still can’t plan for upcoming turns. I’m in the right lane for an upcoming right hand turn. For some reason it goes into the center lane and then signals immediately to return into the right lane ?‍
Maybe they give a dumb version for us Colorado drivers ?‍

Mine does similar things and I live in California. I see people singing praises and I take whatever they say going forward as the opinion of a cult member for the cult leader. I will be overjoyed when (and I do believe it is a when) FSD is great.

 

[ehuna]

I use it everyday. I'm about to use it after I write this post. It works "ok". OK= good enough to be FSD Supervised. But it is nowhere near unsupervised FSD, watching movies or anything like that. I do understand how the technology works and I think by the later part of next year, some form on Level 3 capability in a geofenced area is possible. Maybe on some LA Freeways for example. But I think it will take a bit longer before it is capable of broader unsupervised use cases. I would say ~2027 is more realistic for that.

That seems reasonable, you can bet on a “Tesla robotaxi public release before 2026” on https://kalshi.com/markets/kxrobotaxiout/robotaxi-release

My money is on “before 2026” ?

 

[ehuna]

Mine does similar things and I live in California. I see people singing praises and I take whatever they say going forward as the opinion of a cult member for the cult leader. I will be overjoyed when (and I do believe it is a when) FSD is great.

Nonsense, people are showing their experiences and many record FSD videos like me to prove it - it’s not a cult when there’s data and videos behind to prove it.

You can also go ahead an put your money where your mouth is and bet against a Robotaxi before 2026 here https://kalshi.com/markets/kxrobotaxiout/robotaxi-release

Just like Tesla shorts you will lose your shirt. (shorts?) ?

 

[Darthamerica]

RoboTaxi is possible before 2026. But it isn't going to be a wide release and it will be geofenced. I don't think some people understand how far away v13 is from a customer ready version of whatever SW is in Robotaxi.

 

[SCTesla]

RoboTaxi is possible before 2026. But it isn't going to be a wide release and it will be geofenced. I don't think some people understand how far away v13 is from a customer ready version of whatever SW is in Robotaxi.

Correct 13.2 is far from RT according to Chuck and Dirty, but what we saw at We Robot was mapping and controlled environment. I think we see that in Austin and Palo Alto before 2026, but it will be slow to expand.

 

[Darthamerica]

Correct 13.2 is far from RT according to Chuck and Dirty, but what we saw at We Robot was mapping and controlled environment. I think we see that in Austin and Palo Alto before 2026, but it will be slow to expand.

I agree. People shouldn't get discouraged, they just need to be patient. I think Tesla is making remarkable progress and right where they should be considering the state of the art. One detail I think that gets over looked is how hard it is for Tesla to get quality data for training. Our cars are operating on a static neural network that only changes after and insane about of training. The fleet has to encounter edge cases often enough for the neural network to be able to generalize in a variety of situations. As the cars get better, there will be fewer edge cases to work with for training. I don't expect too many gigantic leaps in performance between releases in the future. Instead it is likely to be little incremental improvements every quarter to 6 months. I think some very limited RT late next year, then the rest of the cars between now and 2027 where Tesla will be able to say they are at Level 3, maybe Level 4.

 

[SCTesla]

I agree. People shouldn't get discouraged, they just need to be patient. I think Tesla is making remarkable progress and right where they should be considering the state of the art. One detail I think that gets over looked is how hard it is for Tesla to get quality data for training. Our cars are operating on a static neural network that only changes after and insane about of training. The fleet has to encounter edge cases often enough for the neural network to be able to generalize in a variety of situations. As the cars get better, there will be fewer edge cases to work with for training. I don't expect too many gigantic leaps in performance between releases in the future. Instead it is likely to be little incremental improvements every quarter to 6 months. I think some very limited RT late next year, then the rest of the cars between now and 2027 where Tesla will be able to say they are at Level 3, maybe Level 4.

Agreed. I've been testing FSD since Oct 2021. It's always one step forward, 2 steps back (except v12 which was 2 steps forward and 1 steps back), but it's progressed. 10.3.1 could barely make a left turn and 9 out of 10 drives would try to kill you. 3 years later 2 out of 10 drives try to kill you, but that's massive progress and it's difficult.

 

[HaulingAss]

This is a lot of YMMV. I live in a rural town in Pennsylvania, and I’ve had full self driving package with my model S for as many years as they’ve offered it, and now with my cyber truck, and I can tell you with certainty that it has never performed well in our town. And yes, the roads are digitized, we’re not that rural.

it is inconsistent, makes decisions that are potentially dangerous, does not follow the speed limit accurately, has problems staying in a safe corridor on unmarked Road, just to name a few.

That's nothing! I live in a rural town in Washington State and the HUMANS here are inconsistent, they make decisions that are potentially dangerous, they don't follow the speed limits accurately and they have trouble staying in a safe corridor.

Because FSD is competing with humans for safety, the bar is very low indeed!

 

[ehuna]

That's nothing! I live in a rural town in Washington State and the HUMANS here are inconsistent, they make decisions that are potentially dangerous, they don't follow the speed limits accurately and they have trouble staying in a safe corridor.

Because FSD is competing with humans for safety, the bar is very low indeed!

Exactly!

Just yesterday, FSD was driving us around and it stopped at a red light. This guy on a motorcycle passed us on the right and completely burned the red light!

Tesla FSD doesn't drink and drive, doesn't text and get distracted, and doesn't get road rage - that's thousands of lives saved right there!

 

[Woodrick]

I’m more comparing it to the “HD Maps” that are needed for LIDAR based solutions - these are called “maps” but they are way more than that, with details on elevation and every object in a given route.

Tesla FSD does not need that - and for example it worked flawlessly in a private road and parking lot in Bend, Oregon, just a couple of days ago.

Lots of misunderstanding going on here.
LIDAR doesn't mean that you are using HD maps. To a degree, any solution that looks at its environment (LIDAR or Vision) shouldn't need HD maps. HD maps allow you to NOT have to determine where the lines are.


And Tesla does use HD mapping, not the maps that are like Googles, but it uses vision to create HD occupancy maps of the area around it.

If you look at any of the occupancy maps that you see a Tesla drawing, it becomes pretty obvious that it vision can "draw" the environment not only as good as LIDAR, but even better. Tesla's occupancy maps draw detailed 3-D renderings of the world around it.

Sponsored