FSD Beta Button in the UK?

[GRiLLA]

As Tesla have sold FSD to people in the UK, and have used regulations as the reason that it isn't very functional, they are going to find it hard to not provide new capabilities as legislation changes.

As the government defined that Level 3 type self driving will be allowed on motorways at under 37mph later this year (which I think is based in UNECE guidance so also relevant to the EU) it would seem that Tesla will inevitably support it. It would seem that the groundwork is being put in place with the cabin camera now detecting attention (the new regs require that the driver is awake and able to take over in 10s) but also require that the driver is kept entertained e.g. with movies, emails etc. So while we may not be getting the 'city streets' self driving, I think we will see some future vision of simple motorway driving where you don't need to be alert and putting pressure on the wheel. Maybe Tesla would use that as a prompt to move to the new rewrite/vision only codebase so we would get incremental benefits elsewhere.

Clearly single lane at < 37mph isn't going to be very unless you are commuting the M25, but it is a trial ahead of the next step being suggested as allowing 70mph. To me this will be far more useful that city streets driving, it's not really been my need or expectation that my car will drive everywhere while I'm closely monitoring it, that's lower value than L3 on long trips will be to me.

 

[JupiterMan]

According to Teslarati and a post from @greentheonly , Tesla is testing FSDBeta in EU right now.

Tesla FSD Beta is currently testing in Europe: hacker

It appears that Tesla’s efforts to bring its Full Self-Driving Beta to foreign territories are underway. The initiative seems to be part of Tesla’s efforts to expand the reach of its advanced driver-assist system, which, together with its expected open Supercharger program, should help the...

www.teslarati.com

 

[GeorgeSymonds]

I'm wouldn't be surprised if somebody in Europe had a car with it to see what happens, but thats a long long way from it becoming available here anytime soon. Teslarati take any rumour and try and turn it into a revenue generating click stream however tenuous the story.

 

[pdk42]

I’d like to be proved wrong, but I think there is a lot of work needed to make the FSD “city streets” system work in the UK. Not only is RHD an issue, but signage, road features, rules of the road, etc will severely complicate things - and that’s without UNECE regs adding to the pain. Personally, I‘d be quite happy to see motorway/dual carriageway/highway working well and ignore the rest. It’s on long, boring roads like that where FSD has most appeal for me.

 

[JupiterMan]

I think there is a lot of work needed to make the FSD “city streets” system work in the UK. Not only is RHD an issue, but signage, road features, rules of the road, etc will severely complicate things

watching the Tesla AI day I see it completely the other way around.... even mini-roundabouts... watching the yielding calculations and monte carlo simulations it plans ahead of time is amazing.

disclaimer: I'm an optimist!

 

[pdk42]

watching the Tesla AI day I see it completely the other way around.... even mini-roundabouts... watching the yielding calculations and monte carlo simulations it plans ahead of time is amazing.

disclaimer: I'm an optimist!

Well, I hope you’re right !

 

[GeorgeSymonds]

watching the Tesla AI day I see it completely the other way around.... even mini-roundabouts... watching the yielding calculations and monte carlo simulations it plans ahead of time is amazing.

disclaimer: I'm an optimist!

Making a good attempt at driving and actually driving are wildly different things. Musk is banking on "orders of magnitude statistically safer than a human" but that's not really going to wash on its own. His metric is on accidents (and his current figures are more than a little dodgy), but a self driving car sat waiting to pull onto a roundabout for 4 hours because the traffic is busy and the car doesn't want to chance it isn't going to be a success for any user. Any disconnect of the system is also a failure even if it doesn't result in an accident.

I really don't understand peoples desire for "almost self driving but isn't" which is the best we're going to get for a good while and all the beta, beta 2, limited beta or beta of a beta software is. Until I can forget about driving for an hour or so while the car cruises down the motorway on level 4 autonomy, I will still have to pay complete attention, and if I'm paying attention then what we have now is good enough (if they can get rid of the phantom nraking). Even level 3 if and when they ever get there just means I can send a text message while driving but little else, I still need to be ready within 10 seconds. "Cool tech" that doesn't do anything after the initial play with it becomes pretty much pointless (I have a house full of cool tech I never really use so I know this), and if its less capable than you as a driver, or slower to act or more hesitant, or gets a speed limit wrong and sits crawling at 40 in a 60 (which Ap did to me the other day) then we'll just over ride the damn thing. Somebody wake me up in 4-5 years when level 4 on motorways might be a reality.

One day it will be great. I applaud the ambition. But there's too many false promises and expensive mistakes along the way at the moment

 

[Dilly]

Interesting development on a local road I thought I’d test out.
there’s a school and a 30mph limit.
last week new 20mph signs appeared either side of the school, which are operational when the yellow lamps, incorporated in the sign, flash.
on TACC, car either completely Ignored them or was gazumped by the 30mph sign about ten yards further on. I didn’t feel any braking or hear any over speed bongs, so I assume the former.
City streets FSD is going to have to go some to sort that out!

 

[gangzoom]

V10 visualisation look really impressive, how long do we think we will need to wait to see it implemented in the UK .

 

[MrBadger]

To be honest, I don't really care about City Streets, but do want them to sort out Highways. If they had put a fraction of the effort of City Streets into Highways, it would be pretty much there. I just hope the City Streets and Highways vision stack merge sorts all the outstanding issues, then push out a localised version that could give us Highways Level 4.

 

[Durzel]

I’d be happy just to see the V10 visualisation, to feel like things were actually moving over here. I think we’re easily a year, probably 2 from seeing this here.

 

[goRt]

To be honest, I don't really care about City Streets, but do want them to sort out Highways. If they had put a fraction of the effort of City Streets into Highways, it would be pretty much there. I just hope the City Streets and Highways vision stack merge sorts all the outstanding issues, then push out a localised version that could give us Highways Level 4.

Highways will flow from the code underlying city streets. Musky posted up the other week that he'd just realised the NN needed training for freeways.

 

[MrBadger]

Yea, I am hoping that the merge of City Streets and Highways vision stacks will bring big improvements, but I have hoped that before - third time lucky at least...

Musky posted up the other week that he'd just realised the NN needed training for freeways.

I thought he said was that Highways was feature complete and greater than 99.999999% (might be a few 9's out) of no injury miles? Maybe he just meant that it was all hard coded and that on highways around Palo Alto during office hours its pretty robust, but I think that I've needed to intervene on quite a few occasions to prevent an issue so probably well less than 99.99% no collision miles had I not intervened.

 

[GeorgeSymonds]

They say a lot of things, often contradictory.

I have a simple take that investing in visualisations isn’t investing in self driving. To beautify what went before has nothing to do with helping the car drive, all it does it offers those that get excited by the visuals something to get excited by and is potentially a form if smoke and mirrors. The car doesn’t care for a visual, it will be using matrices of data or vector plots etc. internally. So the question is whether this is actually better underlying data or just a chunk of change spent on smart rendering to keep the interest alive.

 

[Dilly]

I can’t help but want to turn off the visualisations. Yes they are very pretty but serve only as a distraction for the driver and entertainment for the passengers.
They don’t serve any practical purpose. Might be different if they warned of an impending phantom brake! But, I’m willing to be proved wrong.

 

[gangzoom]

I have a simple take that investing in visualisations isn’t investing in self driving.

Tell that to Waymo whom are spending $$$$$$ on HD mapping and multiple sensor suites. You are right been able to 'see' isn't the same as driving but if you cannot see properly you certainly cannot drive. Tesla appears to be getting close all the 'seeing' aspects needed for driving with cameras alone, this is a massive advantage over the likes of Waymo.

The actual driving is where Dojo is going to make a difference. Tesla currently cannot process all the thousands of hours of video to train/teach the NN how to actually drive, so instead is still using humans to program a large part of actual routing/driving.

Google has shown with AlphaGo Zero NNs work best with minimal human instructions, so in theory once Dojo is up and running, the path planning (essentially driving) will take a big step forwards.

FSD Beta 10.0 is only a glimpse of whats coming and it looks great. Any Tesla owner should be excited about having a more developed version of this code stack pushed to their cars at somepoint in the future.

 

[Artiste]

I admire your optimism but I certainly don’t share it. When it comes to FSD the Musk has a depressing history of massively over promising and under delivering. I prefer to judge him by his past performance. The next iteration of hardware or the next software release is always going to be the answer. Now apparently Dojo is going to solve everything. The latest visualisation will get a lot of people very excited, and you can expect to see loads of fanboy videos on YouTube over the next few days, but in reality is this really a big step forward for autonomy?

 

[GeorgeSymonds]

Tell that to Waymo whom are spending $$$$$$ on HD mapping and multiple sensor suites. You are right been able to 'see' isn't the same as driving but if you cannot see properly you certainly cannot drive. Tesla appears to be getting close all the 'seeing' aspects needed for driving with cameras alone, this is a massive advantage over the likes of Waymo.

The actual driving is where Dojo is going to make a difference. Tesla currently cannot process all the thousands of hours of video to train/teach the NN how to actually drive, so instead is still using humans to program a large part of actual routing/driving.

Google has shown with AlphaGo Zero NNs work best with minimal human instructions, so in theory once Dojo is up and running, the path planning (essentially driving) will take a big step forwards.

FSD Beta 10.0 is only a glimpse of whats coming and it looks great. Any Tesla owner should be excited about having a more developed version of this code stack pushed to their cars at somepoint in the future.

You’ve completely misunderstood what I’ve said
The car needs data to understand the road, that doesn’t need to be represented in a form that humans understand and look pretty, it needs to be in a form that can be applied to the software that controls the car. They have spent a lot of effort making it look pretty for humans.. the question is why? We don’t drive using that visual, it’s really of zero benefit to us, so it’s potentially just to give the illusion they know what the surroundings are when they have no extra information than they had before.

 

[MrBadger]

I can’t help but want to turn off the visualisations.

I think you might be able to, or at least, reduce the real estate it takes up.

 

[Durzel]

I admire your optimism but I certainly don’t share it. When it comes to FSD the Musk has a depressing history of massively over promising and under delivering. I prefer to judge him by his past performance. The next iteration of hardware or the next software release is always going to be the answer. Now apparently Dojo is going to solve everything. The latest visualisation will get a lot of people very excited, and you can expect to see loads of fanboy videos on YouTube over the next few days, but in reality is this really a big step forward for autonomy?

I've seen exactly one feature added to FSD in the ~1.5 years I've owned the car, the Traffic Light Stop and Go thing, and I bought in March 2020 so it's not like I was one of the first people to get a Model 3 in the UK.

Dojo could change everything, and also cure cancer, but even if it does I suspect we'll not know about it in the UK for at least a year, more likely 2, 3 or even more. There are already rumblings about HW4 and the potential for more cameras or relocations of existing ones.

I can blame regulations, but the fact remains that we are far behind the States in terms of the pace of development, and even they are plodding along to some extent. Elon promised an open beta button in March for them, which as we all know never happened. If he is that far off promises to Tesla's home market, imagine how far we are away when UNECE regulations are in the mix, etc.