[UK Only] Single Stack / Highways

[MrBadger]

Single stack, ie object detection, perception and visualisation from FSD City Streets beta is finally released in limited numbers.

For avoidance of confusion, this does not add FSD City Streets beta functionality adds the low level improvements from that to enhance experience over regular highways (rip) autopilot, EAP, highways FSD and, speculating here, hopefully pave the way to reintroduce functionality lost with the removal of USS.

Whilst not officially available in UK yet, thought a thread to discuss/speculate it’s impact in UK market.

 

[MrBadger]

Sounds like some good improvements over regular highways autopilot.

I’m getting the impression that car is more confident.

For those with EAP/FSD, reports of quicker auto lane changes - the 5 or so second delay before lane change commences now around 1-2 seconds. Hopefully this will lead to less lane changes being aborted mid sequence due to timeout, especially those that occur once far has already started to move and doesn’t quite make it to the other lane in time.

Visualisations look better too - good response time even on MCU 2, richer level of detail and some more visual hints on what the car is thinking, such as why it may be slowing. Hopefully for some, it will be accompanied by the (from FSD city streets beta functionality) ability to resize the visualisation panel - so all visualisation or all map if that’s what you want.

 

[WannabeOwner]

For those with EAP/FSD, reports of quicker auto lane changes

The Chuck Cook videos I've seen have it charging down the turn-off-lane (on "motorway") past miles of queuing traffic, merge-request blinker-on all the way, and then barging in at the end Slight exaggeration ... but.

resize the visualisation panel - so all visualisation or all map

Chuck commented that resize visualisation to 2/3rds would be nice

For those not familiar with Chuck Cook here's his first video after installing the single-stack FSD Beta V11.3 (he's done 3 more since)

 

[WannabeOwner]

In Chuck's latest there is a rogue tire rolling down the highway (fell off the back of a pickup, seen a bit further on, by the looks of it). Its not in Chuck's lane, but FSD visualisation does show it

 

[MrBadger]

The Chuck Cook videos I've seen have it charging down the turn-off-lane (on "motorway") past miles of queuing traffic, merge-request blinker-on all the way, and then barging in at the end Slight exaggeration ... but.

That sounds like unconfirmed lane change which we dont have. I was referring to instigating/confirming a lane change then the car waiting 5 seconds before it did anything - then timing out because it didn't complete things quick enough (and people blaming the regs) Sounds like that wait is down to a couple of seconds now.

 

[Fursty Ferret]

The wait has always been down to a couple of seconds. The issue is that the UNECE regulations mandate both a minimum (3s) and maximum waiting time (5s), along with a maximum completion time for the lane change (9s). It's a baffling decision and it's this that leads to the car not changing lanes (because it predicts the maximum time will have elapsed before it completes).

This is why the car will aggressively abort a lane change and swerve back, even though a car might already be moving into the lane you're vacating. There doesn't appear to be any science behind the arbitrary time limits. Seems like it would make more sense if it said "you mustn't force a car in the lane you're moving into to decelerate more than x kph, or at a rate exceeding y ms^2, not pointless bloody time limits.

My personal suspicion is that we won't see single-stack in the UK for a long time. We have an existing system that mostly-works and requires zero ongoing cost and development. The new occupancy network may run the safety side of things in the background, but I doubt we'll see it doing any driving.

 

[GeorgeSymonds]

The wait has always been down to a couple of seconds. The issue is that the UNECE regulations mandate both a minimum (3s) and maximum waiting time (5s), along with a maximum completion time for the lane change (9s). It's a baffling decision and it's this that leads to the car not changing lanes (because it predicts the maximum time will have elapsed before it completes).

This is why the car will aggressively abort a lane change and swerve back, even though a car might already be moving into the lane you're vacating. There doesn't appear to be any science behind the arbitrary time limits. Seems like it would make more sense if it said "you mustn't force a car in the lane you're moving into to decelerate more than x kph, or at a rate exceeding y ms^2, not pointless bloody time limits.

My personal suspicion is that we won't see single-stack in the UK for a long time. We have an existing system that mostly-works and requires zero ongoing cost and development. The new occupancy network may run the safety side of things in the background, but I doubt we'll see it doing any driving.

The UNECE regulations are perfectly workable as every other make of car that does it demonstrates. 6 seconds to change lane sufficiently for the car to think it’s now in the new lane is a long time. I also never had the issue, but then I never indicated in anticipation of a change coming up, if you wait until there’s a gap, indicate, the car indicates a couple of times and changes lane. Never failed. Indicate before there’s a space the car times out. Highway Code also says do the former, indicating isn’t a request to others, it’s information to tell them you’re doing it.

As for single stack, I’ve always conceptually thought the way this all works is there is a spacial engine which works out where everythIng is, and with SS they now also understand the trajectory better for those items. This is what they project into the screen. You can then layer planning on top of that. If they can improve the spacial side, everything thereafter should improve. The alternative is the spatial layer is buried into the planning with no clear separation, I guess that might help specific calls but also seems restricted and can lead to conflict (different scenarios calculating what’s actually happening differently).

So my hope is SS will come, we’ll get the graphics, the lane and speed functionality will improve in operation, even if the feature count doesn’t improve.

 

[Fursty Ferret]

6 seconds to change lane sufficiently for the car to think it’s now in the new lane is a long time. I also never had the issue, but then I never indicated in anticipation of a change coming up, if you wait until there’s a gap, indicate, the car indicates a couple of times and changes lane. Never failed. Indicate before there’s a space the car times out. Highway Code also says do the former, indicating isn’t a request to others, it’s information to tell them you’re doing it.

Do you actually have EAP? Unless there's a gap sixteen miles long the car won't move into it. Literally the only way to make it work on anything but an empty motorway is to flick the indicator on, wait for the timeout to elapse, and then attempt again while the car letting you out wonders WTF you're doing.

 

[WannabeOwner]

I was referring to instigating/confirming a lane change then the car waiting 5 seconds before it did anything

Chuck did a couple of manual-overrides for lane change (just by indicating) and the car moved promptly ... but I presume he would have already checked that lane was clear, so no reason for the car to delay.

I don't remember seeing a USA video of EAP doing a lane change ... which might clarify if the current delay thingie is a UK/UNECE problem, or also exists on USA EAP ... but looks like it is quick on FSD Beta

Unless there's a gap sixteen miles long the car won't move into it

Unhelpful

Literally the only way to make it work on anything but an empty motorway is to flick the indicator on, wait for the timeout to elapse, and then attempt again while the car letting you out wonders WTF you're doing.

not my experience (I have EAP on one car and FSD on the other), and seems not to be @GeorgeSymonds experience either

I have no idea what makes cars behave differently (as seems to also be the case, for example, with phantom braking)

 

[Mullermn]

Do you actually have EAP? Unless there's a gap sixteen miles long the car won't move into it. Literally the only way to make it work on anything but an empty motorway is to flick the indicator on, wait for the timeout to elapse, and then attempt again while the car letting you out wonders WTF you're doing.

This is an over exaggeration but it's definitely true that the current NoA lane changing struggles with anything other than sparse traffic.

Having watched quite a few of the videos of the new stack over the last day or two I think it will bring notable improvements to NoA. The gap finding and acceleration/deceleration look much improved. Currently NoA sucks at overtaking because it pulls out in to the faster lane before it begins accelerating, and then accelerates far too slowly. If they sorted that behaviour out that on its own would make a massive difference.

I think it's fairly likely we will see it over here. Tesla don't like duplicated effort and maintaining a separate software stack for non-US markets would be extra effort. I think it's also plausible that the occupancy network stuff is a foundation for various other functions in the car beyond just self driving - eg, actual matrix headlight functionality.

 

[MrBadger]

The wait has always been down to a couple of seconds. The issue is that the UNECE regulations mandate both a minimum (3s) and maximum waiting time (5s), along with a maximum completion time for the lane change (9s). It's a baffling decision and it's this that leads to the car not changing lanes (because it predicts the maximum time will have elapsed before it completes).

I believe those times are the time frames needed to start undertaking the lane change itself, ie crossing the lane marker/divider. They are not about the time required to wait until starting any lateral movement (ie start moving towards the lane marking) - that time is minimum 1 second.

ie. You hit indicator to start lane change. Car must wait 1 second. Car can then start moving towards the lane marker/divider. It must then start crossing that lane marker/divider line no sooner than 3 seconds, no later than 5 seconds and do the initial lateral movement and lane crossing as one manoeuvrer.

That sounds all pretty reasonable to me. Presently the car sometimes seems to wait some unreasonably (for the circumstances) large amount of time before even starting the manoeuvre. It may well take a couple of seconds of lateral movement before you start crossing the lane marker/divider.

Just my interpretation of the regs though.

 

[WannabeOwner]

the current NoA lane changing struggles with anything other than sparse traffic

I don't remember signalling for EAP to move into a tight space ... I don't tend to drive like that anyway, so I would usually to wait for a reasonable gap ... so my expectation may be more easily met than others

Does the Mad Max setting exist over here (not used it, and can't remember off hand if it is available for this situation)

 

[GeorgeSymonds]

Do you actually have EAP? Unless there's a gap sixteen miles long the car won't move into it. Literally the only way to make it work on anything but an empty motorway is to flick the indicator on, wait for the timeout to elapse, and then attempt again while the car letting you out wonders WTF you're doing.

I had EAP on my last car and covered 15k miles in it.

Our other car is a BMW which has it and that also works fine, all the time, so this is down to poor implementation by Tesla if the only conditions you can get it to work are the ones you describe, and not the regulations.

 

[Dilly]

Do you actually have EAP? Unless there's a gap sixteen miles long the car won't move into it. Literally the only way to make it work on anything but an empty motorway is to flick the indicator on, wait for the timeout to elapse, and then attempt again while the car letting you out wonders WTF you're doing.

It’s odd that some people have a lot of trouble. Mine will pull into the same size of gap that I would and do likewise pulling back in. I have lane change set on mad max.

 

[Mullermn]

I don't remember signalling for EAP to move into a tight space ... I don't tend to drive like that anyway, so I would usually to wait for a reasonable gap ... so my expectation may be more easily met than others

Does the Mad Max setting exist over here (not used it, and can't remember off hand if it is available for this situation)

I'm not talking about physically tight spaces, it's more tight in that the timing of the change needs to be correct. An example would be when you're in the middle lane and need to get in to a left hand lane for an upcoming exit. The threshold where the car will struggle is much lower than for a human, who will either speed up to find an upcoming gap (then slow down to slot in) or (safely) slow down to slot in to one behind. Currently NoA will just drive past maintaining it's 'safe for the lane I'm in' speed but never finding a gap that it's able to commit to.

Edit: I'm intrigued that some people don't seem to have this issue. Now I'm wondering if my problem might go away if I fiddle with some of the Autopilot settings. I'll give it a go.


I would actually like it if they implement the setting that Chuck Cook commented on where you can tell NoA to self drive but just maintain it's lane unless you tell it otherwise. Currently you either have to have the car nag you for permission to change lane or turn NoA off completely and just fall back to TACC.

 

[MrBadger]

I would actually like it if they implement the setting that Chuck Cook commented on where you can tell NoA to self drive but just maintain it's lane unless you tell it otherwise. Currently you either have to have the car nag you for permission to change lane or turn NoA off completely and just fall back to TACC.

That functionality only makes sense if you have unconfirmed lane change does it not? We have to confirm a lane change with indicator stalk so car isn't going to change lanes of its own accord. So whether NoA suggests the lane change and you confirm by stalk, or you manually instigate it via indicator stalk, its still its going to maintain its lane until you signal/confirm by means of the indicator stalk.

 

[Mullermn]

I had EAP on my last car and covered 15k miles in it.

Our other car is a BMW which has it and that also works fine, all the time, so this is down to poor implementation by Tesla if the only conditions you can get it to work are the ones you describe, and not the regulations.

I went looking for some videos of the BMW system because I don't really know what the standard of the other auto manufacturers' 'Autopilot' systems is. Anyway, this is the first result I got that isn't a puff piece about an upcoming L3 system:

Clearly not everyone thinks it's flawless.. Some of those results are quite old though so maybe that's out of date (still top on Google though).
What's the name of the current/best (but actually available) BMW system? I am genuinely interested to see Autopilot in context of the other systems that are out there.

 

[Zilla91]

Those screenshots are about lane keeping, which a lot of people seem to hate irrespective of what brand it is. It has nothing to do with the actual act of automated lane changes.

 

[Mullermn]

OK, that's fine. But what is the equivalent BMW system then?

I'm not pushing an agenda here, it's just that I only know the car I drive which is the Tesla (plus two other vehicles that don't have this level of ADAS). People always pop up and say that Tesla have lost their advantage, other cars are just as good at it these days, etc, so I want to learn a bit more about the competing systems.

 

[Mullermn]

That functionality only makes sense if you have unconfirmed lane change does it not? We have to confirm a lane change with indicator stalk so car isn't going to change lanes of its own accord. So whether NoA suggests the lane change and you confirm by stalk, or you manually instigate it via indicator stalk, its still its going to maintain its lane until you signal/confirm by means of the indicator stalk.

With the current system you lose part of the feedback about what autopilot is doing because the UI is given over to the lane change it wants to make, plus it sticks the blind spot camera over a chunk of the screen. If AP gets hung up on changing to a faster lane in order to gain a small bit of speed then that can end up being in the way for ages. Plus if you condition yourself to routinely ignore that alert you're going to miss a 'change lane to follow route' request one day.

It's a relatively minor gripe, but if the option exists to just have it not change lane unless I tell it to that would be an improved experience for me. Either that or if the lane changing intelligence advances to the point that it always makes the same decision I would that'd be good too - but I suspect that is unrealistic for reasons of personal driving style if nothing else.