FSD on Cybertruck Performs Significantly Worse than on Model 3, S, X (and is missing features)

[carsly]

Mine has a chronic tailgating problem and will run over center medians when making left turns. Training on visual data instead of rules just takes a lot of visual data input and Cybertruck sensor positioning and control systems - especially with four wheel steering and a long wheelbase - require different calibration. It would almost have been easier in 12.x which was rules-based to simply write a few hundred to thousand lines of code to deal with the differences, but in this new world we can't quite do that. Give it time, use it when you can safely and report interventions - that will help it get better, quicker.

Frankly the fact that FSD even exists on such a new vehicle platform in and of itself is quite an achievement. Dumb Summon would be a nice addition though, and shouldn't be terribly difficult. Sometimes I have to jostle cars on my driveway pad and being able to simply pull forward and backward would allow me to get another vehicle out of my garage (Model Y, share the in-garage charger with CT) without having to drive the CT.

Sponsored

 

[TruckGenio]

Since the FSD update, I have been 'testing' FSD out nearly every day. However, I have to frequently disengage due to harsh/ late stopping, unpredictable behavior in intersections, and riding the left side of the lane to the point of hitting a plastic barrier. This has been so disappointing that I decided to no longer use FSD on the truck.

Yet, since the update I have also had to chance to use FSD on the 3, X, and S. The experience on each has been significantly better. Driving is much smoother and actions are more natural. Furthermore, you can customize the driving style on each where you cannot on the cybertruck. Visualizations are also better and more detailed.

What is going on here?? Does the cybertruck run on an entirely different dataset or program? No excuse for FSD being this subpar only on the cybertruck

It’s terrible and if the cybertruck had autopilot I would not use FSD. I think that’s why they didn’t give the cybertruck autopilot.

 

[DAE1]

Cybertruck has a lot of learning to do. when you disengage, make sure to hit the mic and give a quick explanation for why you disengaged. If everyone was sure to do that, I bet they’d learn a lot quicker vs less reports.

I second this! It’s good to leave those recordings so they can improve. I try to use FSD everyday but it usually gives me false alerts thinking I’m not paying attention, so I just disengage, leave a note, and drive normally. Hope they fix it. I haven’t been able to use it much.

 

[DaBagBoy]

What is going on here?? Does the cybertruck run on an entirely different dataset or program? No excuse for FSD being this subpar only on the cybertruck

Patience young grasshopper, you do realize a lot of us waited 3-6 months to even get the first CT version of FSD working. Having driven FSD Beta for 5 years it will get better, like all the other cars have, it is far from "perfect" and clearly has a lot of "issues"....

You just have to accept today and hope for a better tomorrow!!

 

[petercyber]

Since the FSD update, I have been 'testing' FSD out nearly every day. However, I have to frequently disengage due to harsh/ late stopping, unpredictable behavior in intersections, and riding the left side of the lane to the point of hitting a plastic barrier. This has been so disappointing that I decided to no longer use FSD on the truck.

Yet, since the update I have also had to chance to use FSD on the 3, X, and S. The experience on each has been significantly better. Driving is much smoother and actions are more natural. Furthermore, you can customize the driving style on each where you cannot on the cybertruck. Visualizations are also better and more detailed.

What is going on here?? Does the cybertruck run on an entirely different dataset or program? No excuse for FSD being this subpar only on the cybertruck

I 100% agree with you, it feels so unsafe. This needs to be heard by Elon soon.



Thank you

 

[petercyber]

I’m a little confused why Cybertruck can’t use the same data and account for different dimensions/ camera distances

Right?

 

[petercyber]

I have more than once disengagedFSD because it turns to sharp to the left and cut into cross lane too much. Couple times was going to miss my exit

Same here

 

[M0unt41nm4n]

I’m a little confused why Cybertruck can’t use the same data and account for different dimensions/ camera distances

Many reasons. The CT is bigger and heavier as well as has different widths, etc. Its not unlike going from a small car into a F350. You kind of have to relearn stopping distances, how/where to make your turns, understand inertia, and different type of speeds under varying conditions. Whenever someone drives my F350 for the first time, its funny because its like driving with a grandmother. They are nervous and overly cautious. They always say "Wow, that feels so different". Hence you can't just use software that works great on a smaller/lighter vehicle on something so much bigger and heaver. It needs to be trained.

Ford has the same issues with their AI. They have Bluecruise on the F150 which works rather nice. But they are still training to get it on Super Duty trucks (F250/F350) because of the differences that I just explained.

 

[igs]

You kind of have to relearn stopping distances, how/where to make your turns, understand inertia, and different type of speeds under varying conditions.

lol FSD is vision based it has no idea what inertia is nor does it care.

 

[M0unt41nm4n]

lol FSD is vision based it has no idea what inertia is nor does it care.

Haha... so... I assume you are in the business of AI and write AI code? ? *NOT*

The vision determines the inertia/weight by measuring it against braking distances. How do you think AI learns to play video games? How do you think humans are able to play video games like driving games? You think its because they feel inertia and weight or because they can determine stopping distances correlated with the vision of the games?

Here ya go @igs ... play this and tell me how you determine inertia and weight:

http://moonlander.seb.ly/

Hint: Its all visual!

 

[igs]

Haha... so... I assume you are in the business of AI and write AI code? ? *NOT*

Well you clearly are not and I'm not clicking on your spam link bro. You're confusing the old FSD that had models and vectors with NN which Elon has stated is just pixels in controls out.

 

[M0unt41nm4n]

Well you clearly are not and I'm not clicking on your spam link bro. You're confusing the old FSD that had models and vectors with NN which Elon has stated is just pixels in controls out.

Spam link? hahaha...ok. No I don't put my profession in my "spam link"

But I'll give you a hint...Yeah bud, I actually write AI models, prepare data features, and train AI for a living. Yes my expertise is in Python utilizing pytorch and Tensorflow with a host of other APIs utilized for that sort of stuff. Yes, it's what paid my mortgage and for my Cybertruck. ?

So wait... lets be sure we are on the same page here and speaking the same language bro, "pixels in and controls out" means visual in to do the controls. ? The Tesla vehicles are now trained with 300K lines of C++ code removed and uses only a neural network (that's AI - Read your release notes). Didn't I say it was all visual? ?‍

 

[65SoYoLO]

I just discovered that if you gently steer the truck back to the middle of the lane It will say that way for a little bit if it starts to veer to one side steered back to the middle I think that is how it is learning instead of disconnecting it.yrmv

 

[igs]

But I'll give you a hint...Yeah bud, I actually write AI models, prepare data features, and train AI for a living. Yes my expertise is in Python utilizing pytorch and Tensorflow with a host of other APIs utilized for that sort of stuff. Yes, it's what paid my mortgage and for my Cybertruck. ?

FSD doesn't use any of that. Sorry bud.

 

[M0unt41nm4n]

@igs

Here ya go my man!

https://www.qwak.com/post/pytorch-vs-tensorflow

TensorFlow is a powerful tool for building and training neural networks, but it’s also more complex than many other deep learning frameworks.It is used for major projects across the world and is used by companies such as Airbnb, Google, Uber, Tesla, and more.

https://codersbarn.com/does-tesla-use-pytorch-or-tensorflow/

When deciding between PyTorch and TensorFlow, Tesla opted for PyTorch as its primary deep learning framework. This choice reflects Tesla’s commitment to a flexible and open-source platform that allows for rapid development and customization. By selecting PyTorch, Tesla gains the freedom to experiment with cutting-edge machine learning techniques and tailor models to their specific needs without being tied down by rigid structures.

OMG...

FSD doesn't use any of that. Sorry bud.

Nooooo.... they don't... ?

Sponsored