Roundabouts and FSD

[madcal]

Guys - your not selling this FSD to me. There must be some upsides to persuade may to shell out £5k+!!

Well, I did because:

1. I want them to continue developing it
2. I could afford to
3. But mostly it annoyed me when on the motorway I had to disengage TACC every time I overtook someone (often). Now with FSD and NOA enabled I can happily sit at my cruise speed, agree with the car by nudging the indicator when it wants to overtake then do it again to get back in. No jerky disabling of TACC to overtake. I also like that it reminds me about motorway junctions and tries to turn off because I am easily distracted...
4. Summon; what a joke in the UK. Unless you sit the key fob on the bonnet you can forget about anything other than novelty 2 meters forward and back to show off to your friends
5. Self parking; I can park, never used it.

I live in hope that it will get better, I drove a Tesla in the US with it a month or so ago and it was miles better. Over here it seems nobbled.

 

[Tony Hoyle]

FSD at the moment is fairly dumb lane following, no intelligence with road signs etc. at all (it's not clear it will ever have this as that is patented). Although even with that limited behaviour it is really useful on motorways (and I wouldn't do without it).

Apparently though it's much better than the competition, which must be truly horrific..

Your 5k goes for future development, and to get excited when elon announces that full self driving will be here next year, again.

 

[EV-Fixme]

Had a look on StreetView ... it's not a roundabout.

lol maybe not, looks like a GIANT round about to me on the map.

 

[dc121gw]

Had a look on StreetView ... it's not a roundabout.

What tipped you off? That it passes under and over an 8 lane freeway?

Maybe Bugs Bunny lives in the Corona Mall

 

[RealEvil_UK]

Have had it on 2 loan cars now. If it was £1000 I'd buy for my 3 but for what they charge it's a complete joke.

For those that did buy it, if you could have trialed it for even 1 week would you gave signed up and paid the money? If so... Would genuinely love to hear why.

 

[Battpower]

Have had it on 2 loan cars now. If it was £1000 I'd buy for my 3 but for what they charge it's a complete joke.

For those that did buy it, if you could have trialed it for even 1 week would you gave signed up and paid the money? If so... Would genuinely love to hear why.

FSD envy. Compare to cosmetic surgery.

 

[GeorgeSymonds]

A few simple questions to those who have had a Tesla for a while:

When do you think Teslas in the UK will read speed limit signs, including variable ones?
When do you think they will stop at stop lines and give way at give way lines when TACC is engaged?
When do you think TACC will have the ability to adhere to speed limits as they change?

The ability to report map errors with a button press would be good - TomTom had this years ago! However the advertising puff suggested this wouldn’t be needed in a Tesla because it would be picked up from camera data being fed back to ‘HQ”!

Well.. cars built before 2017 can read speedlimits... they use a third party system which Tesla have just about caught up with on basic features but not speedlimits. That said, its not 100% reliable but it still does it. It also doesn't do anything with that data other than put it in the display, it would be great if it adjusted TACC down when it drops Its said to be a patent issue why Tesla don't now have it but i suspect this is more a convenient excuse.

Stop signs and traffic lights is said to be on its say but initially it will be more a warning.

For me the primary issue is reliability of any of these features. I'd prefer it to do less but more reliably than Tesla currently trying to push out a thousand features none of which work when some cars still seem to get phantom braking near a bridge. And the less but more reliable is probably a good way to describe the original AP1 system.

Its also a shame that the Tesla marketing and sales machine talks about features when they're first thought of rather than when they're ready

And I can't accept the idea of "I bought FSD because I am happy to fund development". Tesla are the second most valuable automotive company in the world. I can't see anyone ever saying they'd buy a £5k option on a Ford Mondeo so Ford can develop it, and then wait. And some people have been waiting for 3 years for FSD and its still nowhere to be seen.

 

[KennethS]

And I can't accept the idea of "I bought FSD because I am happy to fund development". Tesla are the second most valuable automotive company in the world.

They weren't when I bought mine in July 2019. In fact, at the time, the Tesla shorts were pushing the narrative that the company was on the verge of collapse. That said, I did not buy FSD because I was trying to save the company or help fund its development. As my first Tesla and a gadget geek, I wanted the full package and knew I was paying more than was justifiable on a purely economic analysis. I do like the summon feature (we have a tight garage and it is handy for pulling the car in and out without having to struggle exiting the car), and the self parking. I went in fully aware of the limitations of the system and its quirks (which are many).

 

[GeorgeSymonds]

They weren't when I bought mine in July 2019. In fact, at the time, the Tesla shorts were pushing the narrative that the company was on the verge of collapse. That said, I did not buy FSD because I was trying to save the company or help fund its development. As my first Tesla and a gadget geek, I wanted the full package and knew I was paying more than was justifiable on a purely economic analysis. I do like the summon feature (we have a tight garage and it is handy for pulling the car in and out without having to struggle exiting the car), and the self parking. I went in fully aware of the limitations of the system and its quirks (which are many).

That’s fine but summon, lane change etc were all part of AP1 which was about £2k when I got in my first car in 2015. You’re obviously entitled to spend your money how you wish, but plenty have bought with the expectation based on Teslas promise that something will come along soon. There was a famous tweet, I think 2 years old, may even be 3, where Musk said FSD would deviate from EAP within “3 months, 6 months definitely” or words like that. They’ve now taken half of EAP like summon and put it in FSD.

The question is would you have paid if they told you you’d get nothing extra for 2 years and you can buy then at more or less the same price?

 

[LukeT]

For me the primary issue is reliability of any of these features. I'd prefer it to do less but more reliably than Tesla currently trying to push out a thousand features none of which work when some cars still seem to get phantom braking near a bridge.

This.

But the features list sells, I suppose. Saying "it does the same things much better than before" would be everything to the existing owner who isn't buying and nebulous at best to the non-owner who might be buying!

For me it would feel like a much much better system if, for example:
- lane change worked quicker, didn't involve a deceleration, and didn't have a habit of aborting
- it looked ahead enough, combined with mapping info, or whatever required, to get trajectory and speed right through a corner
- it looked ahead enough to smooth its decelerations in traffic rather than leaving them so late
- and of course if the (pretty constant in UK winter) 'blocked or blinded' issue was cleared up, but that feels like hardware...

Basically it works in principle for say 80% of my journey, and I use it for that, but has these issues for all of that. So tackling most of the issues in the 80% would have a much greater positive effect on my experience than the presumably harder job of creating the necessary features to autopilot the remaining "hard to do" 20%. I'm quite happy taking control for the odd roundabout.

 

[Battpower]

For example:
- lane change worked quicker, didn't involve a deceleration, and didn't have a habit of aborting
- it looked ahead enough, combined with mapping info, or whatever required, to get trajectory and speed right through a corner
- it looked ahead enough to smooth its decelerations in traffic rather than leaving them so late

All of that.

And a big yes to predictable hard fantom braking that for me has been evident exiting tunnels and going under gantries especially with large lane direction panels etc.

Actually, the whole issue of inconsistent behaviour is far worse than missing a feature completely. It makes identification and reporting somewhat fraught, even more so when behaviour varies from one country to another in in different weather conditions.

And the whole 'last minute decision' feel to ap is far from inspiring. If the car can know that in 50 yards the speed limit will drop to 20mph and your current speed is 35, why, why bother trying to accelerate to 40mph for those few yards, especially since the car doesn't tend to react to the 20 limit for another 50 yards or so.

All the ap / fsd features will have to start reading ahead more in order to feel in any way natural and reassuring.

 

[MrBadger]

All the ap / fsd features will have to start reading ahead more in order to feel in any way natural and reassuring.

Absolutely. Near the top of my 'why' (does it do that) list is that when exiting a motorway on NOA it leaves indicating until it has already started to pass the offramp. It knows its coming off as the sat nav shows its going to take the exit, but why leave indicating to far too late.

I'm sure some people will argue that it is wrong expecting the car to behave in a human like way when it can make much later decisions, but we are going through a transition phase for the occupants and other road users expect a car to behave in the same way as a normal car and follow established Highway Code rules and recommendations.

 

[tsh2]

AP is probably in data/edge case collection and validation against human drivers right now. There is no need, and potentially some risk, in making incremental improvements. Particularly in the non-A-road scenarios I think there needs to be a high level of confidence before the 'acceptability' is improved. As soon as AP is able to navigate anything above the most benign roundabout scenario, there needs to be a failsafe (stop if situation not handled).

 

[Battpower]

I'm sure some people will argue that it is wrong expecting the car to behave in a human like way

Probably, but they wouldn't be justified! Fortunately we still live in real human centred society and for way into the future except maybe on certain designated routes it will be the driver who has responsibility for their cars. For as long as we have real drivers on the roads, all vehicles have to follow conventions that allow safe, consistent and predictable interactions.

 

[Battpower]

AP is probably in data/edge case

Yes. And probably a difficult hard edge (edges) , given that regulatory bodies are going to keep Tesla well away from all obvious edges, while Tesla has to explore every facet of every edge without accidentally falling over any.

That development process and being part of it in a small way is probably the biggest reason for having FSD in its present state.

 

[gangzoom]

For those that did buy it, if you could have trialed it for even 1 week would you gave signed up and paid the money? If so... Would genuinely love to hear why.

Because am a geek.

I didn't even buy the original AP when we ordered our car in 2016. Like all Tesla owners I had plenty of opportunity to try loaners with AP, both AP1 and than early AP2, both were pretty crude.

The change for me was the introduction of 'Tesla vision', and actually all of a sudden AP2 code seems to be progressing.

When Tesla dropped the price of FSD from £9k to £5k I took the view actually it probably was the right balance of cash versus geek.

Our AP2 car is miles better on the AP front when we took delivery, and anyone thinks AP1 is still better than AP2+ cars are deluded.

Will we ever get FSD, who knows, but when an iPhone costs over £1k and MacBook Pro over £2.5k when speced up, £5k for me isn't that much to spend on the biggest gadget I own .

 

[Battpower]

I just had a concern thinking about this.....

Since the idea of driving schools using FSD cars would seem counter productive, I wonder where else the neural network training data is being gathered to give rise to these driving characteristics?

 

[Durzel]

They weren't when I bought mine in July 2019. In fact, at the time, the Tesla shorts were pushing the narrative that the company was on the verge of collapse. That said, I did not buy FSD because I was trying to save the company or help fund its development. As my first Tesla and a gadget geek, I wanted the full package and knew I was paying more than was justifiable on a purely economic analysis. I do like the summon feature (we have a tight garage and it is handy for pulling the car in and out without having to struggle exiting the car), and the self parking. I went in fully aware of the limitations of the system and its quirks (which are many).

This is mostly where I sit with it.

It's manifestly not worth £5800, but I sortof equate it to the same as the Touch Bar on the current MacBook Pros. I find it mostly useless, and it hasn't really moved on massively since it was launched, but at the same time it kinda feels like its party piece.

As illogical as it sounds - having the ability to summon your car, as neutered as it currently is (although that hack supposedly makes it better), and the full fat self driving, feels like it would be conspicuous in its absence on an otherwise "tomorrow's world" car.

 

[madcal]

I will fully agree it was crazy expensive at over £5,500 just to get the NOA to work on motorways. But like some of the others I am a geek and love the latest stuff. Ironically it was driving a friends one with FSD that made me purchase it for mine a week after getting mine delivered without it.

And just in case you were wondering I am not a Tesla-can-do-no-wrong type. I am sooooo irritated with how terrible after sales service is. If you buy one just hope you have no issues.

 

[MrBadger]

Someone elsewhere posted this link used on a Tesla recruitment page. Thought that I would pass it on in context of this thread.

30 second video https://www.tesla.com/sites/default/files/images/careers/autopilot/network.mp4

Gives some good insight into what AP is seeing, even on city streets with junctions - whether it is actually driving is not known. Not sure if it is pre or post (another) autopilot rewrite that is apparently near completion. But other shots from the recruitment page look right up to date as they show code for stop sign recognition.

[edit]
2 things that I notice most of interest.
First, that lights (stop, indicators) on vehicles seem to be of little interest. And second, how quick it picks up stop signs - I'm wondering if its getting these from known locations its so fast, but then, its also highlighting them precisely too. Of note, it also picks up an equally occluded pedestrian. So either a mega cheat, or very impressive.
I also like the driving around the puddle at first intersection - wonder whose driving? Very positive if its AP - there is hope for potholes and speed bumps!