FSD rewrite will go out on Oct 20 to limited beta

[CurtisVL]

Select group of safe, careful and expert drivers, in USA I presume.

In EU we only get privileges to pay for FSD, and not the possibility to use any of the functions...

Having said that, the UK is still within the EU for now and we might be getting hands-off self driving as early as next year for use on highways only.

 

[rebelfocus]

Having said that, the UK is still within the EU for now and we might be getting hands-off self driving as early as next year for use on highways only.

The current regulations are not EU specific, they are UNECE which represents 17% of the worlds population and 56 member states. The UK isn't getting away that easily unless it wants to go rogue on the UN as well (wouldn't surprise me).

 

[CurtisVL]

The current regulations are not EU specific, they are UNECE which represents 17% of the worlds population and 56 member states. The UK isn't getting away that easily unless it wants to go rogue on the UN as well (wouldn't surprise me).

Of course, but it feels right now like the UNECE regulations are being followed by the EU mainly and ignored by most of the rest of the world, so I could see the UK moving away from them if they didn't change!

 

[mikes_fsd]

UNECE which represents 17% of the worlds population and 56 member states.

The fact that a single bureaucracy has the power to cripple 17% of the populations ability to get a driving aid that works in the other 83% of the population is the mindboggling part.

rogue on the UN

But you choose to focus on the fact that a population that might want out from under such BS.

TF is wrong with the world.

 

[Bill Price]

I'm sure they need a UK tester, right? I'm always happy to help out.

OK...........limited to one UK tester and one US tester?

 

[Mardak]

Are you an expert and careful driver?

Interesting that there's this additional qualification. Either the "FSD beta" has serious regressions in some corner cases or there's new functionality, e.g., turning at intersections.

I wonder if this early release is limited to those with FSD software capability (as FSD hardware is definitely required). Even without turning through intersections, basic Autopilot of staying within a lane and other features like speed limit sign detection (which requires hardware but not software) should be much improved too.

 

[diplomat33]

Interesting that there's this additional qualification. Either the "FSD beta" has serious regressions in some corner cases or there's new functionality, e.g., turning at intersections.

I wonder if this early release is limited to those with FSD software capability (as FSD hardware is definitely required). Even without turning through intersections, basic Autopilot of staying within a lane and other features like speed limit sign detection (which requires hardware but not software) should be much improved too.

I imagine the FSD beta will be pretty close to "feature complete". It will almost certainly be able to make turns at intersections and probably be able do a basic commute from home to work completely on AP. I think that is why it will require "expert and careful" driver since Tesla wants a responsible driver who can properly supervise AP through intersection turns etc... If AP is handling making turns at intersections with traffic, you will definitely want a good driver who can really pay attention to what AP is doing.

Yes, I think it will most likely require the FSD software package since this is the "real FSD" that Elon has been promising.

 

[Spaceosaurus]

Interesting that there's this additional qualification. Either the "FSD beta" has serious regressions in some corner cases or there's new functionality, e.g., turning at intersections.

If what Elon says is true the new FSD beta will be a total rewrite, so they can't make any assumptions on safety from what the current AP can do. They're basically regressing to 0 public hours of real world testing from the millions they have currently. I'll bet their confidence in Palo Alto FSD capability is super high, everywhere else... not so much.

 

[Dutchie]

Timing is so that will have something to brag about in the conference call, which happens the next day.

 

[Snuffysasa]

It'll be interesting to see how they determine expert and careful drivers.

Fred over at Electrek claims that drivers will still have to keep their hands on the steering wheel. Is there any confirmation of that? I think it would be awkward if the car needs to make a 90 degree turn and it turns the steering wheel quickly.

Well for sure they still require the driver to be watching at all times and stay engaged... and since they don't have another way to measure this, keeping hands on wheel is pretty much only option...

It will be like before where it will nag every so often. You can keep your hands on the wheel as the vehicle does a 90 degree turn and let the wheel slide against your fingers. (I do this all the time when testing 90 degree turns with autonomous cars at work)

Though, they probably won't ensure hands are on the wheel continuously thought the turn... but only at the start... you will likely need to confirm to proceed through the turn, possibly by putting your hands on the wheel before the turn starts. Just like they did with lane changes.

 

[mikes_fsd]

If what Elon says is true the new FSD beta will be a total rewrite, so they can't make any assumptions on safety from what the current AP can do. They're basically regressing to 0 public hours of real world testing from the millions they have currently. I'll bet their confidence in Palo Alto FSD capability is super high, everywhere else... not so much.

Not quite, the individual NN's that are run against the separate camera feeds would still have all the "experience" or heritage that they've accumulated prior to the rewrite.

But they are taking out all the hardcoded stitching of results across cameras and probably a bunch of other hard coded stuff.
So definitely still a risk in driving policy, not perception.

 

[Spaceosaurus]

Not quite, the individual NN's that are run against the separate camera feeds would still have all the "experience" or heritage that they've accumulated prior to the rewrite.

But they are taking out all the hardcoded stitching of results across cameras and probably a bunch of other hard coded stuff.
So definitely still a risk in driving policy, not perception.

Even if they're using the same ML models (which is doubtful, given the things Elon has said about how model calculations are done from multiple views, which is different than today), it would be the equivalent of taking the tires off your old car, putting them on your new car, and then saying your new car has 30,000 miles worth of driving. Nope, one part of your car has 30,000 miles, but certainly not the whole car as a unit. The model doesn't run in isolation, it's executed and arbitrated by a bunch of other code which hasn't seen real world testing.

 

[Florafauna]

The fact that a single bureaucracy has the power to cripple 17% of the populations ability to get a driving aid that works in the other 83% of the population is the mindboggling part.

But you choose to focus on the fact that a population that might want out from under such BS.

TF is wrong with the world.

FSD works perfectly well in the UK (and presumably the rest of the EU) - within limits.

FSD, however, is massively US centric. Lots of straight roads. FSD will take a long time to cope with the single track roads that make up 30% of the roads round here - 50 miles from London. Will it reverse to a lay-by?

Even the main roads connecting towns are single lane each way and can have 90 degree bends (or more), FSD doesn't even slow down.

I'm loving the FSD journey - that's why I got it. However it is FAR from being able to cope with anything over here that isn't a multi lane highway.

Nothing to do with the EU or UN.

 

[Ryan H]

Well for sure they still require the driver to be watching at all times and stay engaged... and since they don't have another way to measure this, keeping hands on wheel is pretty much only option...

It will be like before where it will nag every so often. You can keep your hands on the wheel as the vehicle does a 90 degree turn and let the wheel slide against your fingers. (I do this all the time when testing 90 degree turns with autonomous cars at work)

Though, they probably won't ensure hands are on the wheel continuously thought the turn... but only at the start... you will likely need to confirm to proceed through the turn, possibly by putting your hands on the wheel before the turn starts. Just like they did with lane changes.

I think this could be a requirement for beta testers:
Tesla hacker reveals what driver-facing camera is looking for - Electrek

 

[mikes_fsd]

Even if they're using the same ML models (which is doubtful, given the things Elon has said about how model calculations are done from multiple views, which is different than today), it would be the equivalent of taking the tires off your old car, putting them on your new car, and then saying your new car has 30,000 miles worth of driving. Nope, one part of your car has 30,000 miles, but certainly not the whole car as a unit. The model doesn't run in isolation, it's executed and arbitrated by a bunch of other code which hasn't seen real world testing.

No, it is nothing like that.
A NN trained to recognize road traffic data that has been trained on billions of miles of real world traffic data is still a NN that has that knowledge.
It is like putting a pro driver into a new indy car.
They have the experience, but the feel of the new car the driver will still need to get accustomed to.

 

[ItsNotAboutTheMoney]

I imagine the FSD beta will be pretty close to "feature complete". It will almost certainly be able to make turns at intersections and probably be able do a basic commute from home to work completely on AP. I think that is why it will require "expert and careful" driver since Tesla wants a responsible driver who can properly supervise AP through intersection turns etc... If AP is handling making turns at intersections with traffic, you will definitely want a good driver who can really pay attention to what AP is doing.

Yes, I think it will most likely require the FSD software package since this is the "real FSD" that Elon has been promising.

More likely they don't want any headlines to be the result of dumb stunts by dumb ... stuntspeople.

 

[SOULPEDL]

Well for sure they still require the driver to be watching at all times and stay engaged... and since they don't have another way to measure this, keeping hands on wheel is pretty much only option...

They likely already have ability to read your attention level with the cabin camera, but only if absolutely necessary as last resort I'd think.
In a turn, I guess it will free wheel (and will take another level of raw nerves for some at busy intersections). But initially will be more of the same requiring driver confirmation before executing a maneuver. Like you must use your turn signal and pedal pressure at the light for confirmation to proceed. Then HANG ON!

Looking forward really. It's the whole reason we ordered FSD in 2016. It's been a pleasure the whole way, and plenty more fun ahead.

 

[Spaceosaurus]

No, it is nothing like that.
A NN trained to recognize road traffic data that has been trained on billions of miles of real world traffic data is still a NN that has that knowledge.
It is like putting a pro driver into a new indy car.
They have the experience, but the feel of the new car the driver will still need to get accustomed to.

By your logic the vNew FSD would lead to the same behaviors as the vOld FSD. An existing, tested model executes an existing, tested action. If either of those statements is false, either the model or the action being untested, then by default the chain is untested.

You're absolutely right that the model will benefit in object recognition by all the billions of miles of real world data. That model will be further enhanced by the ability to recognize and track objects across multiple camera views, as Elon's said publicly. That improved model has not been tested by billions of real world miles.

Similarly, you have logic executed based on those models. Today, you will see phantom breaks due to any number of interpreted signals from that model. Elon has suggested that no longer happens. That new, changed behavior has not been tested by billions of real world miles.

This is in contrast to today, where adding one additional piece of functionality doesn't invalidate the entirety of AP, because the change is supplemental and iterative, vs. a fundamental replacement.

If you work in software, which it sounds like you do, you know that the con of a rewrite is that it has the potential to introduce far more bugs than a single, incremental update to an existing codebase.

 

[Redbrick]

I wonder ...if fsd has been redeveloped from ground up....what does this mean to those who did not buy fsd? In otherwords, is tesla going to maintain a 'have' and 'have nots' version? (the resources..yikes) ...or will it be like windows 'home' vs 'pro' version? Personally I can't wait since I have purchase fsd, but nervously excited that we're all lab rats...hopefully not perceived as disposable.

 

[ChrML]

This is in contrast to today, where adding one additional piece of functionality doesn't invalidate the entirety of AP, because the change is supplemental and iterative, vs. a fundamental replacement.

If you work in software, which it sounds like you do, you know that the con of a rewrite is that it has the potential to introduce far more bugs than a single, incremental update to an existing codebase.

This is true, however as you probably know 75% of modern software development is about testing frameworks, writing test scenarios and collecting/labeling test data. The actual code is only 25% of the work.

Done correctly you can do fairly substantial rewrites, and validate it against your testing framework and it should work quite well out of the box. Eventual loopholes found during early testing is used to improve the tests.

Even though Tesla doesn't collect billions km of driving, they've hopefully built an impressive automated testing suite from 4 years of production experience.