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.
I think self driving lorries will be resisted quite heavily as it's putting an entire industry out of business. It'll be more a political than business decision.

Same way they keep investing in coal and oil even though it can only ever be short term.. those workers vote.
 
I think self driving lorries will be resisted quite heavily as it's putting an entire industry out of business. It'll be more a political than business decision.

Same way they keep investing in coal and oil even though it can only ever be short term.. those workers vote.
Can delay it for a while, will not be able to delay it forever. Also when you've got globalisation, if your the country with expensive human drivers or I can put my factory in a country with cheap AI based driven lorries then you might well lose out on that factory.

You are right it's a big risk, not sure in UK but in US the largest employment sector is transportation. Honestly though it's likely to take a lot of our jobs, if not all of them regardless of what you do over time. We'll be surplus to business and need a universal income though otherwise there's no one to buy the goods and services companies make anyway.
 

I wouldn't get too excited, these kinds of amendments pass through GRVA all the time. It's unlikely the first text will be voted through, so that means more data gathering/redrafting for subsequent meetings, plus once it gets through that stage it has to go to the full WP29 which can take several months even if they don't make any changes.

It's a slow bureaucratic process. If we see anything practical in under 12 months I'd be surprised. More likely a couple of years.
That would be great but I'm not sure I follow.

Surely if Ford BlueCruise & Mercedes can already do hands free on motorways in Europe, it means the regulation already allows it?

Why would we need a new piece of amendment to the legislation specifically for Tesla to do it?
 
That would be great but I'm not sure I follow.

Surely if Ford BlueCruise & Mercedes can already do hands free on motorways in Europe, it means the regulation already allows it?

Why would we need a new piece of amendment to the legislation specifically for Tesla to do it?
Have to say this baffles me too, is the main restriction, other than ensuring the driver is paying attention, is the g-force around corners?

If that's the case then VW managed that with their predict ahead on my ID.3 years ago.

It slowed me down for corners and also roundabouts and junctions.
 
It's more about unattended lane changes.. the limits for the hands free are quite strict and it's not surprising tesla don't go for it.

g force is a red herring.. it's perfectly sensible to slow down as you go around corners, just as any driver would. AP is incapable of it.
 
AP was capable of it, at least until the UNECE hobbled the steering wheel turn rate to that of less than a typical human driver.

However, AP officially is designed for fast-flowing A-spec roads free of pedestrians and cyclists, similar principles behind the UNECE decision.

Hopefully their trip to Michigan will help persuade them to be less problematic towards self-driving progress.
 
There is no streering wheel turn rate.

The rules simply say you have a maximum g force going round corners, and the limits are set at what comfortable for a human passenger.

AP has always been incapble of this. Instead it just aborts and heads for the nearest hedge.

Lots of other manufacturers systems handle it just fine. Tesla is *years* behind the curve on this.. not least because they basically abandoned AP development for the US only 'city streets' stuff.
 
There is no streering wheel turn rate.

The rules simply say you have a maximum g force going round corners, and the limits are set at what comfortable for a human passenger.

AP has always been incapble of this. Instead it just aborts and heads for the nearest hedge.

Lots of other manufacturers systems handle it just fine. Tesla is *years* behind the curve on this.. not least because they basically abandoned AP development for the US only 'city streets' stuff.
So my understanding was right.

I let my ID.3 manage the speed around some right corners, those with an advisory speed limit, and it stuck to the advisory limit fairly closely.

My M3 will just happily let go mid corner, my ID.3 never did that, it just slowed a 'little' more than I would have, but still not drastically slower.

That's what you want in autonomy, certainly in the early days, caution with capability.
 
  • Like
Reactions: Tony Hoyle
My M3 will just happily let go mid corner, my ID.3 never did that, it just slowed a 'little' more than I would have, but still not drastically slower.

That's what you want in autonomy, certainly in the early days, caution with capability.
The Tesla autopilot system could easily do those corners too - we know that the same builds did fine in the USA. Its just the UNECE insisted that certain G-forces could not be exceeded, and Tesla implemented this to deactivate the system rather than to slow the car and reduce the G loads to within UNECE constraints (like VW).

It's a shame that Tesla did not spend more time implementing UNECE compliant code to help us more over here, but its fair to say that they had bigger fish to fry in attempting to solve FSD as opposed to coding for our regulators.

We see similar today, where Tesla is not keen on reprogramming supervised FSD to abide by UNECE rules, which would be quite problematic and time constraining, possibly requiring a re-code of the rules based V11 rather than the better end to end V12.
 
The Tesla autopilot system could easily do those corners too - we know that the same builds did fine in the USA. Its just the UNECE insisted that certain G-forces could not be exceeded, and Tesla implemented this to deactivate the system rather than to slow the car and reduce the G loads to within UNECE constraints (like VW).

It's a shame that Tesla did not spend more time implementing UNECE compliant code to help us more over here, but its fair to say that they had bigger fish to fry in attempting to solve FSD as opposed to coding for our regulators.

We see similar today, where Tesla is not keen on reprogramming supervised FSD to abide by UNECE rules, which would be quite problematic and time constraining, possibly requiring a re-code of the rules based V11 rather than the better end to end V12.
Sorry, but I don't really follow why it's remotely challenging to follow the UNECE imposed limits, FSD must have similar to get the car around a corner safely, they're just slightly more conservative values.

At some point both systems must use mapped data to determine the turn radius of each corner on the road network and assign a max speed allowing for a max g-force, simple maths compared to recognising cars, people, bins, which it's much better at than determining a car exists.
 
FSD must have similar to get the car around a corner safely, they're just slightly more conservative values.

At some point both systems must use mapped data to determine the turn radius of each corner on the road network and assign a max speed allowing for a max g-force, simple maths compared to recognising cars, people, bins, which it's much better at than determining a car exists.
You are missing that FSD is now an end-to-end Neural Net. There is no coding/setting of g-force limits. It is all trained by example videos showing how it should drive. As far as requiring map data, FSD doesn't require it, though it is an input to the NN when it is available. (From what I have seen, it will drive on brand new, non-mapped, roads without an issue. But it obviously can't use them for navigation.)

To get a different output you would need different example videos and to train an entirely separate NN to meet the UNECE limits. (Maybe Tesla can come up with a different way to do that, but I thought I heard that UNECE was set to increase the g-force limits to actual reasonable values soon, so it might not be necessary.)
 
You are missing that FSD is now an end-to-end Neural Net. There is no coding/setting of g-force limits. It is all trained by example videos showing how it should drive. As far as requiring map data, FSD doesn't require it, though it is an input to the NN when it is available. (From what I have seen, it will drive on brand new, non-mapped, roads without an issue. But it obviously can't use them for navigation.)

To get a different output you would need different example videos and to train an entirely separate NN to meet the UNECE limits. (Maybe Tesla can come up with a different way to do that, but I thought I heard that UNECE was set to increase the g-force limits to actual reasonable values soon, so it might not be necessary.)
It really wouldn't be difficult to overlay g-force limits, this has already been done by competitors obviously.

The limits set by UNECE aren't altogether unreasonable, I've let my old car do corners and it's not overly slow.

Autonomy should be cautious, it shouldn't be able to exceed speed limits for example.
 
If an end to end neural net is that inflexible they'll have to completely retrain it for each country anyway.. one more rule isn't special.
They would not be able to train the models to abide by UNECE rules (not only g-forces, but lane changes, indication timing etc) as practically no human driver drives like that.
 
They would not be able to train the models to abide by UNECE rules (not only g-forces, but lane changes, indication timing etc) as practically no human driver drives like that.
If UNECE has a unique way of indicating before changing lanes...it won’t be long before other drivers recognize when the system is employed (just like you always know if the car in front is using adaptive cruise control)...so other drivers will start ignoring the lane changing manœuvre knowing the system will abort
 
  • Like
Reactions: JupiterMan
The limits set by UNECE aren't altogether unreasonable,
The current UNECE rules on motorway overtaking, from time to start change procedure, gap to vehicle behind, speed of lane change, to time out if incomplete, are IMHO completely unreasonable, to a point where I have difficulty seeing Robotaxi's being practicable in UNECE territories if those rules persist.

G-force limits are also far too low.

UNECE currently lives in an ideal world where all drivers are nuns and saints, and comes up with these strict rules on how things should happen. But the real world does not work like that.

Its similar in some ways to how Tesla used to code FSD up to V11, where much was heuristics / rule based - they have learned that you cannot programme a set of rules/instructions for all the roads - things are far more complicated, which is why V12 is doing a much better job - you learn from the drivers on those roads, not by a rule book formulated at a conference in Geneva.

I am of the opinion that UNECE should give up on the the strict rules they are implementing, and instead adopt a strategy based on the outputs / real events of the autonomous vehicles - perhaps some kind of driving test for autonomous vehicles, similar to that we see now for a human driver. if they pass this then they are set free, where fleet data/safety can then be monitored by the regulators and improved as necessary,
 
It's more about unattended lane changes.. the limits for the hands free are quite strict and it's not surprising tesla don't go for it.

g force is a red herring.. it's perfectly sensible to slow down as you go around corners, just as any driver would. AP is incapable of it.
I feel like with so many saying autopilot is incapable of this and I generally don't use it on the roads where this is needed. However, it has slowed down for corners that I was pretty sure it wouldn't be able to take and then kind of taken them. When I say kind of taken them it's not always successful but it has been able to do it OK.

Feels like they really could cover this with just a bit mors work, just not sure they care to invest their time developing a code base that is vastly out of date and should be binned off sooner or later. Honestly I'm kind of surprised they haven't already done it.
 
If UNECE has a unique way of indicating before changing lanes...it won’t be long before other drivers recognize when the system is employed (just like you always know if the car in front is using adaptive cruise control)...so other drivers will start ignoring the lane changing manœuvre knowing the system will abort
There's nothing unique about indicating before changing lanes.. it's what you're supposed to do.

Autonomous cars *should* be restricted to drive perfectly. Humans are crap at driving - the accident statistics show that. Learning off them is IMO the wrong way to go about it.