FSD rewrite will go out on Oct 20 to limited beta

[diplomat33]

Something that I can use is more advanced than something I can't use.

No. Those are two different things. You are confusing availability with capability. One does not necessarily mean the other. If I am able to fly in a two seater propeller plane but I can't fly in a F-16, does that make the propeller plane more advanced than a F-16? No, it does not. I think we would agree that a F-16 is more advanced.

Waymo can be more advanced because it has more capable hardware and software while also being less available. And Tesla's FSD can be less advanced because the hardware and software are less capable while being more available.

 

[powertoold]

Can we assume it needs a map to work?

I don’t think it requires a map to work. However, as a safeguard to consumers, it’s a good idea to only allow activation in a map area. Otherwise, what’s to stop it from being activated in a random field or off-road? FSD is intended to be a robotaxi in mapped areas. I wouldn’t consider this as a geo-fence however.

 

[diplomat33]

I don’t think it requires a map to work. However, as a safeguard to consumers, it’s a good idea to only allow activation in a map area. Otherwise, what’s to stop it from being activated in a random field or off-road? FSD is intended to be a robotaxi in mapped areas. I wouldn’t consider this as a geo-fence however.

Oh the irony.

 

[MP3Mike]

Can we assume it needs a map to work?

It is a regression in this version, since it worked in the prior version. We don't know if it is intentional or not.

 

[firsttruck]

I don't think the rate of improvement is linear though. Right now, improvements are coming fast because FSD is still early. As FSD gets better, solving those rare edge cases, gets harder and harder. Getting a 10% improvement gets harder and harder.

Edge cases are things that occur less frequently in the real world but that does NOT also mean they are harder for the NN to solve. Most might quite easy to solve.

 

[diplomat33]

Edge cases occur less frequently but that does not also mean they are harder for the NN to solve. Most might quite easy to solve.

We don't know that. Some edge cases might be easier to solve, while others might be harder.

But if they occur less frequently, they will be harder to find and it will take longer to gather enough data to solve them. So it will slow down your progress.

 

[powertoold]

Oh the irony.

Considering it as a geofence is like saying Uber is geofenced.

Even if we say it is geofenced, there’s no irony here, as the geofence involves the entirety of the USA mapped roads.

 

[firsttruck]

We don't know that. Some edge cases might be easier to solve, while others might be harder.

But if they occur less frequently, they will be harder to find and it will take longer to gather enough data to solve them. So it will slow down your progress.

1. Not all edge cases are important. A solution might not even be a priority.
2. Seeing more occurrences of a specific edge case may not be needed. One occurrence can provide enough data for a solution.
3. Solution of a specific edge case can solve edge cases not yet encountered.

My point is that there can be many many edge cases that are easy to solve with immediately available data.

 

[diplomat33]

Seeing more occurrences of a specific edge case may not be needed. One occurrence can provide enough data for a solution.

I don't think this is correct. Machine learning requires a certain amount of data in order for the NN to get it right. One occurrence will not be enough data for the NN to solve it.

 

[Bladerskb]

Nope, this is all Tesla. I believe Tesla is able to do magical things with their NN because they own the entire stack. They can use the car's sensors (steering position, accelerometers, odometers, GPS, etc.) to measure all sorts of road characteristics derived from vision. Because they are able to control every aspect of their sensor positioning and angles, focal lengths, etc. etc., along with their huge fleet, they can label large amounts of data using "future" labels to label past images. That's how they're able to achieve other cars' positions and speed despite the cameras being partially blocked by a front car here:

https://twitter.com/teslaownersSV/status/1320098691680665600

Edit: who am I kidding? I'm probably not even scratching the surface of what Tesla is doing to achieve mindboggling NN predictions.

This is the same old hydranet from 2018 with some improvements that Tesla came out with to finally match Mobileye in neural networks deployed. You are telling me that finally having what someone else has always had means you are 10 years ahead and what you are doing is mindboggling and magical?

Green "same hydranet pretty much"

https://twitter.com/greentheonly/status/1301170668428578818

https://twitter.com/greentheonly/status/1321134418300489728

https://twitter.com/greentheonly/status/1318344088253571072

As green said 'The outputs of the hydranets are then ingested into BEV NN(s)"

https://twitter.com/greentheonly/status/1320947823530123265

Has anyone found any potentially FSD-breaking issues with this beta?

At this point, I can only see them solving everything to average human level within the next 1-2 years at most.

Basically, it looks like Tesla has won the race to a widely deployed FSD system. Just my insight.

Let me get this straight. Tesla releases an update which primarily contains software 1.0, classical c++ conventional control algorithms that everyone is using. Infact some others are using way more ML and NN in their driving policy stack. But by using their 2 years old hydranet which they deployed to catch up with Mobileye EyeQ4 from 2017. They won self driving?

Wait what? So

https://twitter.com/greentheonly/status/1301170668428578818

From what I’m seeing of the FSD beta, Tesla is very close to a level 5 system that is nowhere as safe as a human but *is capable* of navigating through all of the routinely driven roads.

So that means that Mobileye has already achieved level 5 back in 2017 and we are not even talking about their EyeQ5 NNs yet.

It's crazy I even think this, but 6-9 months lol.

Do you even know that most lidar classification uses BEV networks? And that BEV networks are industry standard and even andrej said the same thing?

Here's a BEV network from 2018 and 2020 for example

BirdNet: a 3D Object Detection Framework from LiDAR information
https://arxiv.org/pdf/2003.04188.pdf

The old 2018 hydranets outputs literally feds into their BEV networks. Outputs such as detected objects (Cars, peds, cyclist, etc), road edge and markings and path prediction.

Literally Mobileye could use the outputs from their EyeQ4 from 2017 if they wanted to create their own BEV network back then. (Figure #1 contains real world example of their 3d all angle vehice detection)

Even more is the case with their next gen EyeQ5.

Yet to you tying a bunch of networks that Mobileye has had since 2017 into a run of the mill industry standard BEV network and writing some classical conventional control algorithm to you is Level 5 and human level disengagement stats will be achieved in just over 6 month.

This is prime example of willful ignorance.
We already have BMW releasing software that stops and goes at traffic light/stop signs. Which works anywhere (Entire EU, uk, canada, china, US, etc).

This is a control algorithm from a dev team (BMW/ Tier 1 ZF) that isn't anywhere near the cream of the crop.

Yet you think, turning at intersections and overtaking is somehow means Level 5 at safety level near, at or better than human will be achived in 6 months.

EDIT:
Figure #1
NIO

BMW

VW

NIssan

 

[powertoold]

We already have BMW releasing software that stops and goes at traffic light/stop signs. Which works anywhere (Entire EU, uk, canada, china, US, etc).

not yet, we can assess when they actually deploy it and people test it out in various conditions, like direct sunlight.

 

[DanCar]

... You are confusing availability with capability.

No, you are confusing your definition for my definition. My definition is something that I can use is more advanced than something I can't. Something I can't use isn't worth much and very un-advanced.

 

[jeremymc7]

We’re on the 3rd release so far, that I can confirm, and the latest seems to have substantial cosmetic and likely performance improvements.

Start pushing it out to a wider beta group. It’s quite stable, compared to the latest wide public release, so get it out there for more pool reporting / capture which will speed things up even more.
Seems like maybe only a dozen, few dozen, hundred (non employee) people have it.

 

[diplomat33]

Basically, it looks like Tesla has won the race to a widely deployed FSD system. Just my insight.

Not yet. Tesla does not have a widely deployed FSD system. Tesla has not deployed the FSD to the entire fleet yet.

 

[Mo City]

We’re on the 3rd release so far, that I can confirm, and the latest seems to have substantial cosmetic and likely performance improvements.

Start pushing it out to a wider beta group. It’s quite stable, compared to the latest wide public release, so get it out there for more pool reporting / capture which will speed things up even more.
Seems like maybe only a dozen, few dozen, hundred (non employee) people have it.

IMO, they shouldn't push it out to a wider group until the current group is no longer able to generate enough edge cases to keep Tesla busy.

 

[diplomat33]

No, you are confusing your definition for my definition. My definition is something that I can use is more advanced than something I can't. Something I can't use isn't worth much and very un-advanced.

ad·vanced
adjective

1. 1. modern and recently developed.
   "the team developed advanced techniques for measuring and controlling the noise of the submarines"
   
   Similar: state-of-the-art
   
   
2. 2. far on or ahead in development or progress.

There is nothing in that definition about being widely available or useful. You are just making up your own personal definition. Now, I grant you that if the tech is not useful to you, it might not have much value to you. That's fair. But that has nothing to do with how advanced the tech is.

 

[Thp3]

No, you are confusing your definition for my definition. My definition is something that I can use is more advanced than something I can't. Something I can't use isn't worth much and very un-advanced.

Please don’t. We use English to share data. Using your own personal definitions isn’t helpful.

I think your view can be said as
“Something I can use is more helpful than something I can’t”.

valid viewpoint, if I summarized it right, but it’s not about technology, advanced or simple. It’s simply what you have access to at the moment.

 

[P3dStealth]

IMO, they shouldn't push it out to a wider group until the current group is no longer able to generate enough edge cases to keep Tesla busy.

I agree with you but for different reasons.

I don't want tesla to push it out to a bigger group because I'm afraid the regulators will ban it and then target tesla autopilot when someone crashes.

It's far more of a beta than anything else they've previously released and it's far more dangerous. I do not want usa regulators to have a reason to decide my steering wheel torque isn't enough to make sure I'm paying attention.

While a lot of us here are highly aware of the system and it's capabilities... Most people just buy the car and click buttons. Most average consumers don't belong in the beta if the car might decide to drive into a dead end. My girlfriend prob wouldn't even notice till her phone fell out of her hand.

Push it out when it's ready. If they rush this it could hurt us all because the world is watching.

 

[Mo City]

I agree with you but for different reasons.

I don't want tesla to push it out to a bigger group because I'm afraid the regulators will ban it and then target tesla autopilot when someone crashes.

It's far more of a beta than anything else they've previously released and it's far more dangerous. I do not want usa regulators to have a reason to decide my steering wheel torque isn't enough to make sure I'm paying attention.

While a lot of us here are highly aware of the system and it's capabilities... Most people just buy the car and click buttons. Most average consumers don't belong in the beta if the car might decide to drive into a dead end. My girlfriend prob wouldn't even notice till her phone fell out of her hand.

Push it out when it's ready. If they rush this it could hurt us all because the world is watching.

How do you know this is "far more of a beta than anything else they've previously released"? What exactly does that mean? Are you familiar with details of their beta program?

I think Tesla is well aware of what would happen if a general release resulted in a big upswing of accidents. After all, they promote AP as increasing safety which rationalizes them using the public to test beta software. For that reason, my view is "feature complete" FSD will not go general until sometime in 2021.

 

[S4WRXTTCS]

IMO, they shouldn't push it out to a wider group until the current group is no longer able to generate enough edge cases to keep Tesla busy.

Has anyone noticed if the group of testers has started to grow?

I haven't personally noticed much change of it, and maybe Tesla is perfectly happy with the size of the current group for now.

What I hope to see is more geographical areas covered within the US. Like I haven't seen any representation yet in WA state.

Now I am completely against a wide release or substantially increasing the testing size which I don't believe you were either. It needs to be kept small to avoid the whole regulatory crack down, and bad press. But, needs to be big enough to generate useful data.

Plus it needs to get bigger so it seems like less of a marketing stunt, and more of something that's actually going to be released within a reasonable time frame (less than a year).

If they are going to actively geofence it then it does seem a bit plausible. Hopefully someone will try Lombard street again to see if it allows it to be activated.