Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

UK FSD Discussion

This site may earn commission on affiliate links.
FSD Beta 12.3 now getting wider release in the US. Noticed it has already managed to curb one of the Plaid owners’ wheels making a left turn.
Not perfect but honestly I find it extremely impressive and still despite watching a few FSD videos over the years, mind boggling that the car can do all this on its own. V12 does seem to be a big step up from v11 so for once Elon was right to be excited about it. He didn’t take the Beta tag off though, guessing legal wouldn’t let him 😉
 
Without going down the "ha what about the magic roundabout" route, I've noticed these US roundabout videos are pretty simple. Traffic flows are usually very low; lanes are clearly marked and usually 2 wide; usually only "cross" or "T" shaped.

Would like to see a video where it has to wait a while and be incredibly assertive to enter the roundabout. Some examples near me that come to mind
1710674209892.png
1710674298895.png


Maybe it _is_ just training data that's needed, but it also feels like our roads are much less standardised so would need more cases to train against? Even the lights need to be top notch about assigning which lane they are referencing and reading the coloured arrows. In the US it always seems to be "your light is above your lane". Example:
1710674455939.png

These ones here were actually improved recently. Left light is for straight ahead in the two left lanes, the one at red there is for the right turning lane. There used to be a single light there that had aspects like this which I'd never seen before being a single light was both red AND green at the same time:
Untitled Diagram.drawio.png
 
  • Informative
  • Like
Reactions: Roy B and Pink Duck
FSD had to be able to interpret english (or other languages) on road signs too. If a sign says '20 when lights flash' that clearly means don't do 20 at other times. It has to cope with this.. https://c8.alamy.com/comp/H5XARY/trial-50-mph-for-hgvs-on-the-a9-scotland-H5XARY.jpg. It should be reading bus lane signs and working out whether to use that lane based on what time it is..


I get the impression US roads and signage are much simpler.
 
Without going down the "ha what about the magic roundabout" route, I've noticed these US roundabout videos are pretty simple. Traffic flows are usually very low; lanes are clearly marked and usually 2 wide; usually only "cross" or "T" shaped.

Would like to see a video where it has to wait a while and be incredibly assertive to enter the roundabout. Some examples near me that come to mind
View attachment 1028872View attachment 1028873

Maybe it _is_ just training data that's needed, but it also feels like our roads are much less standardised so would need more cases to train against? Even the lights need to be top notch about assigning which lane they are referencing and reading the coloured arrows. In the US it always seems to be "your light is above your lane". Example:
View attachment 1028874
These ones here were actually improved recently. Left light is for straight ahead in the two left lanes, the one at red there is for the right turning lane. There used to be a single light there that had aspects like this which I'd never seen before being a single light was both red AND green at the same time:
View attachment 1028875
I agree the videos I've seen of FSD with roundabouts in the US all seem to have low traffic volumes and it's not trouble at all. However it couldn't do them at all in earlier versions so it's moving forwards nicely. Ours will be harder but if they can train on enough UK based data shouldn't be something that isn't possible. It's just a matter of time and seeing as I'm not convinced they are working on this yet, many years in the future for us.
 
  • Like
Reactions: randompixel
It may be quite some years before it makes it here, if at all. People seem to lament that FSD is currently focussed on the USA. I think it is understandable why.

I do find myself following via YouTube the progress Tesla is making and find it oddly captivating. It seems 12.3 is an evolutionary change and with additional compute power coming online at Tesla HQ perhaps the improvements will be more rapid.
 
  • Like
Reactions: randompixel
FSD had to be able to interpret english (or other languages) on road signs too. If a sign says '20 when lights flash' that clearly means don't do 20 at other times. It has to cope with this.. https://c8.alamy.com/comp/H5XARY/trial-50-mph-for-hgvs-on-the-a9-scotland-H5XARY.jpg. It should be reading bus lane signs and working out whether to use that lane based on what time it is..

I get the impression US roads and signage are much simpler.
Essential for any & all Trump supporters.
 
12.3 follows usual pattern where people are blown away initially before reality sets in and all the flaws start to show up after a few weeks

That person getting their wheel kerbed and someone bumping their car in a car park shows it’s not fit for purpose as a robotaxi or fully autonomous system and it never will be no matter how many times they keep polishing the turd
 
  • Like
Reactions: Alwayshopeful
12.3 follows usual pattern where people are blown away initially before reality sets in and all the flaws start to show up after a few weeks

That person getting their wheel kerbed and someone bumping their car in a car park shows it’s not fit for purpose as a robotaxi or fully autonomous system and it never will be no matter how many times they keep polishing the turd
The only way we don't get self driving cars is if we wipe humanity off the planet or send ourselves back to the stone ages. Otherwise it might not be coming in "two weeks" like Elon time but it's going to come. Machines will be better than us at pretty much everything, if not everything. It's just simply a matter of time.
 
Last edited:
  • Like
Reactions: Drew57 and KennethS
Next up @FastLaneJB will tell us we will all be driving flying cars in the not too distant future 😂

Do you think Star Wars is an accurate depiction of the future?
Deflect as much as you like ;) I think most of the experts in this field think GenAI will surpass us, it already has in some ways. I've 24 odd years in one career that I think makes me somewhat of an expert in it however I still don't know it all, AI does while also knowing tons about doctors, lawyers, accountants and so on.

Look at it this way. We are born with just basic knowledge on how to breath, drink milk but that's about it. We are always learning but over 16 - 18 years we are training and learning off others. AI is literally does the same but just faster and with far more information than we get to process.

Can it go wrong with bad data? Sure but many kids develop to be quite like their parents because that's the data they get trained on. It's all pretty simple really.

I didn't think it would happen iny lifetime 20+ years ago but we've moved forward so quickly, not sure now.
 
I think most of the experts in this field think GenAI will surpass us,
They would say that, wouldn’t they? 🙄

AI does while also knowing tons about doctors, lawyers, accountants and so on.
Don’t do yourself a disservice, it’s really not as clever as you think it is. It’s pretty easy to trip up something like ChatGPT, whether it’s asking it to write some code and then calling it out when the sample given is doesn’t work or something as simple as asking it to summarise what happened in a random football match from 15 years ago.

AI is literally does the same but just faster and with far more information than we get to process.
Great. When it can do what the human brain can do, let me know. Crunching a load of data doesn’t impress me, nor does the amount of compute power and electricity it uses.