FSD v12.x (end to end AI)

[FSDtester#1]

There are companies that claim it’s possible to safely remotely drive cars.

Home - Welcome to Vay: Leading Company for Remote Driving

Vay is a leading teledriving company that aims to launch a unique, affordable, and sustainable door-to-door mobility service.

vay.io

Sorry Julien, it appears Daniel is smarter than you. It's ok to be #2 though.....

 

[Daniel in SD]

Not sure how that could be legal since that is the same as L0 driving. Seems if in an accident or moving violation the driver would need to be in the car since it is NOT autonomous and the same as any other car being driven by a person. How does the police see the driver's license for instance?

Lots of stuff is legal in Nevada.

 

[JulienW]

Sorry Julien, it appears Daniel is smarter than you. It's ok to be #2 though.....

Just a funny. Used to be a decent runner and in 2 races, ½ marathon and a 10K I managed to came in 4th place (my best placements). 4th place in a race is actually 1st in last place.

EDIT: Weirdly sometimes when I'm running by brain thinks I have FSD engaged and I have to force myself back into reality so I don't hit someone.

 

[FSDtester#1]

Lots of stuff is legal in Nevada.

In Florida also. Most of the videos I have watched and articles I have read state that Florida is one of the only states that have almost no restrictions of any kind on AV's....
Have I done the research to back up this claim, Nope. However where there is smoke, fire usually follows. (Got that Emoti, or is it Emoji?)

 

[gottagofast]

ass will not be part of 12.4?

 

[AlanSubie4Life]

You wouldn't put AP mileage into a FSD usage statistic, even if it used the same NN.

But that is exactly what they are doing. This is extremely well documented. Just look at the increase in usage. First big jump in slope was single stack. Then another big step later, similarly explained by highway miles.

It’s predominantly highway miles in the FSD stats. This is well known.

Just post the prior analysis here. It’ll make it clear.

 

[BBTX]

Among the 20,000 TeslaFi subscribers, about 7,000 (including myself) are stuck on 2024.3.x but have received 5 or more updates to FSD v12, all on 2022.3, up to 12.3.6 . I call us the testers. Many were on FSD beta back when a high Safety Score was required.

I've been assuming (perhaps incorrectly) that the "testers" group we've seen during v11/v12 is basically everyone who got into beta before they finally removed the safety score requirement, and nobody from afterwards. Do we have evidence to the contrary?

Hmmm now that I think about it, they probably had to expand a bit beyond that definition just to include more HW4 cars in the tester set.

 

[sleepydoc]

You better believe it. I have been part of some similar research back in the past as a Time & Motion study engineer, and also as noticed in my own daily driving experience. The underlying motto being "Keep moving" is always better, which is also why automobile assembly lines are always in motion. They never stop at any station.

The safety factor also needs to be considered. Even if accidents are rare they are so expensive that a single accident can dwarf everything else.

 

[sleepydoc]

Apologies if this has been posted before, but I'm trying to figure out if any setting changes can fix this mistake FSD beta keeps running into?

Running 12.4.6 - Real life example just today: Navigating to highway entrance from regular street roads. Two left turn lanes but the one on the right is the correct one to choose as the highway entrance is on the right lane after the turn. Instead, the Tesla purposely goes into the turn lane on the left, completes the turn and then tries to force itself into a lane change to get on the lane to turn into the highway. Why does it do this when it clearly knows that staying in the turn lane on the right is clearly the only realistic option to do?

Is anyone else running into this? I feel like I'm doing something wrong if this is happening. FWIW I live in Savannah, GA. Is it maybe because there's not enough Tesla drivers here for them to correct these mistakes? TYI!

I’ve noticed this more lately but I’ve also noticed that the nav system doesn’t specify which turn lane you should be in (I.e. they are both highlighted/bolded.) if FSD is getting incomplete or poor info from the nav system it will make poor decisions.

 

[sleepydoc]

Absolutely. I never do. Today was an exception simply because it was an empty slow backroad. But I should not do it. I totally agree.

I thought FSD would give you a strike if it saw you on your phone?

 

[sleepydoc]

My point when I posed the question "why use FSD if you need to monitor it anyway" is that there is more value in FSD than just its utility. It has entertainment value as well, which is a big part of why I use it, and why I was against eliminating left turns several pages ago. That's not to say I find it useless. I agree it can make driving in unfamiliar areas easier, which also happens to be when it's the most entertaining to use. I also appreciate the frequent updates which keeps it fresh.

“Why use cruise control when you need to steer anyway?”

 

[enemji]

I thought FSD would give you a strike if it saw you on your phone?

It shows the red hands of death

 

[TK211X]

12.3.6s eagerness to accelerate decisively on poor road surfaces causes self induced take over request due to TCS activating longer than ~1 second.

 

[JulienW]

ass will not be part of 12.4?

Looking not likely and going by Elon's latest posts probably not going to be in 12.5 either. So could be at least 12.6 or so.

 

[willow_hiller]

Not sure how that could be legal since that is the same as L0 driving. Seems if in an accident or moving violation the driver would need to be in the car since it is NOT autonomous and the same as any other car being driven by a person. How does the police see the driver's license for instance?

There's a German company doing L0 remote control that the BBC covered last year:

They started in Berlin, and they've since expanded to the UK.

 

[MP3Mike]

But that is exactly what they are doing. This is extremely well documented. Just look at the increase in usage. First big jump in slope was single stack. Then another big step later, similarly explained by highway miles.

It’s predominantly highway miles in the FSD stats. This is well known.

Just post the prior analysis here. It’ll make it clear.

Sure, I get that, with the V11 single stack that FSD started to be used on the highway and count in the FSD usage stats. But are you saying that people are now going to suddenly turn FSD off and use basic Autosteer/NOA on the highway? Or why would the mileage go down if they didn't upgrade AP/NOA to use the FSD stack?

 

[JulienW]

There's a German company doing L0 remote control that the BBC converted last year:

They started in Berlin, and they've since expanded to the UK.

Strange but I doubt many states would approve this. But regardless I bet my last Tesla will NOT do this. Any takers? This seems so anti-Tesla. They will likely follow the standard L4 autonomous tradition and start with safety drivers. Or if they only make a controller less car they will be L4 with remote assistance off the bat.

 

[cyborgLIS]

“Why use cruise control when you need to steer anyway?”

You're suggesting the value of cruise control is that it spares you the physical fatigue of operating the pedal? But if you're looking for relief while driving, you are probably taking a long highway trip, in which case FSD is a bit overkill compared to any other cruise control +/- lane centering system, and probably even less safe for that matter. You only even need cruise control really, since most of the physical demand comes from keeping your foot engaged in a narrow range. Even more relief would come from massaging seats but you would have to look to a different manufacturer for that.

 

[mgs333]

Very interesting. Obviously Tesla tracks MPI - but never publishes it. Now he claims 5x to 10x improvement - but without actually publishing the data, who knows ?!

Anyway - if true - we should see an impact in the crowd sourced FSD Tracker. From 20 MPI to 100 MPI ?

ps : He says intervention - not disengagement. So, the current MPI is probably around 5.

This seems like terrible news for me personally. I have to intervene with the accelerator every few seconds so for me, my MPI is around 0.1-0.2. While I would welcome a 5-10x improvement, something tells me 12.4 will not be keeping up with the flow of traffic either.

But to lose v11 on the highways (in 12.5) where it (v11) actually goes the speed I want it to and have the car routinely drive under the speed limit on the freeways around here means I would stop using FSD altogether as I've already started to use it less and less on city streets.

 

[dgatwood]

ass will not be part of 12.4?

Oh, it will be a**, all right.