FSD video completely fake?

[NerdUno]

This is EAP, not FSD -- so one cannot comment on how FSD is doing.

We've always associated FSD with map-based navigation and EAP with following the dotted lines. I'm not sure that is necessarily the way this is going to play out. I seriously doubt you're going to load an update and go to FSD cold turkey. Have you forgotten what all of us have endured this past year? It would be like starting all over... again. Not sure Tesla's stock price could take it.

 

[buttershrimp]

The way the lane display bounces around all the time isn't exactly confidence inspiring. Their lane marker detection is clearly crap, they just dampen it enough to prevent the car constantly correcting.

They are a long, long way from FSD.

I'm not sure if you are referring to my video, but there is only one part towards the end of the video where the car slightly veered beyond the lane. and I almost had to disengage. Right after the truck passes on the left at the end of a hill. when slope changes from downward to upward... .otherwise, it stayed perfectly centered... I'm just saying .44 is really improving for me at least.

 

[Kant.Ing]

Tesla is fond of and never hesitates pushing beta features via OTA updates. Still no sign of auto wipers or FSD functionalities in OTAs indicates it's either: 1) impossible to work to some acceptance level even with follow-up OTAs and no one dares to call out hardware/architectural limitation but only to exploit software means or change heads, or 2) they really want to make the feature perfect this time around.

 

[AnxietyRanger]

I would like beta auto wipers.

 

[banned-66611]

This is EAP, not FSD -- so one cannot comment on how FSD is doing.

Look at the FSD video. They are using maps to navigate and simple road following in between turns. If they can't get the road following to work then they can't do FSD.

 

[banned-66611]

I'm not sure if you are referring to my video, but there is only one part towards the end of the video where the car slightly veered beyond the lane. and I almost had to disengage. Right after the truck passes on the left at the end of a hill. when slope changes from downward to upward... .otherwise, it stayed perfectly centered... I'm just saying .44 is really improving for me at least.

Look at the dashboard display. The road bounces around constantly.

 

[croman]

Look at the dashboard display. The road bounces around constantly.

If you think that's bad, where were you for the past year? This is ROCK SOLID compared to the insanity that was being pushed before. The operation of the car matters more than IC lines.

 

[AnxietyRanger]

If you think that's bad, where were you for the past year? This is ROCK SOLID compared to the insanity that was being pushed before. The operation of the car matters more than IC lines.

Is that the definition of Stockholm syndrome?

 

[buttershrimp]

Look at the dashboard display. The road bounces around constantly.

Indeed, when I watch the video, the lines jump around on the display, but I'm pretty sure it's just the car celebrating Young Jeezy.

Also, @croman is right, the car behavior is what matters, but it is an interesting point, I would think the accuracy of the display would be more precise than the performance of the car, but it's not, which is interesting in and of itself.

Also @AnxietyRanger, it's not Stockholm Syndrome, it's battered spouse syndrome.... you have no idea how much my Tesla used to beat me after she's been drinking... she's been sober for months now.

 

[croman]

I think the IC lines jump so much in @buttershrimp 's video because it is getting confused between the blacked out lane lines and the reflectors embedded in the road. It was a lot for it to track and it did its best and luckily the lines didn't turn into a concrete barrier.

 

[AnxietyRanger]

One issue I believe is the fact that EAP as it stands (and as it has been specced as at 4 cameras?) looks the world through a rather narrow box, which is fine when the lane is uniform, predictable and not too wide. In this case the lane and its markings are all over the place.

Second thing is that what we see on the IC obviously is not what EAP "sees", but a graphical approximation/averaging of some parameters coming from AP2, so as the lane "lives", the approximation "lives" too. This is made worse by the fact that EAP only sees a small portion of the road at a time anyway...

So if a lane were to go zig-zag or the lane lines made an unexpected detour, the IC still would display it as a straight line or a gently curving road or lose lane markings for a while, or alternate between a couple of states, but not show a zig-zag on the screen. Partially because EAP would only see a portion of it at a time, but mostly because the IC clearly is not getting such parameters that would allow it to draw a complex lane shape.

So no matter how accurate or inaccurate EAP may be at this time, the IC seems to be always averaging it out. Sometimes this helps give a better-than-real impression, and at other times it makes for a worse impression.

Now, the third thing is the driving algorithm, it probably also averages things out these days and does not follow every momentary stray the NN might suggest.

 

[buttershrimp]

One issue I believe is the fact that EAP as it stands (and as it has been specced as at 4 cameras?) looks the world through a rather narrow box, which is fine when the lane is uniform, predictable and not too wide. In this case the lane and its markings are all over the place.

Second thing is that what we see on the IC obviously is not what EAP "sees", but a graphical approximation/averaging of some parameters coming from AP2, so as the lane "lives", the approximation "lives" too. This is made worse by the fact that EAP only sees a small portion of the road at a time anyway...

So if a lane were to go zig-zag or the lane lines made an unexpected detour, the IC still would display it as a straight line or a gently curving road or lose lane markings for a while, or alternate between a couple of states, but not show a zig-zag on the screen. Partially because EAP would only see a portion of it at a time, but mostly because the IC clearly is not getting such parameters that would allow it to draw a complex lane shape.

So no matter how accurate or inaccurate EAP may be at this time, the IC seems to be always averaging it out. Sometimes this helps give a better-than-real impression, and at other times it makes for a worse impression.

Now, the third thing is the driving algorithm, it probably also averages things out these days and does not follow every momentary stray the NN might suggest.

I think you are absolutely right on this

 

[banned-66611]

One issue I believe is the fact that EAP as it stands (and as it has been specced as at 4 cameras?) looks the world through a rather narrow box, which is fine when the lane is uniform, predictable and not too wide. In this case the lane and its markings are all over the place.

Second thing is that what we see on the IC obviously is not what EAP "sees", but a graphical approximation/averaging of some parameters coming from AP2, so as the lane "lives", the approximation "lives" too. This is made worse by the fact that EAP only sees a small portion of the road at a time anyway...

So if a lane were to go zig-zag or the lane lines made an unexpected detour, the IC still would display it as a straight line or a gently curving road or lose lane markings for a while, or alternate between a couple of states, but not show a zig-zag on the screen. Partially because EAP would only see a portion of it at a time, but mostly because the IC clearly is not getting such parameters that would allow it to draw a complex lane shape.

So no matter how accurate or inaccurate EAP may be at this time, the IC seems to be always averaging it out. Sometimes this helps give a better-than-real impression, and at other times it makes for a worse impression.

Now, the third thing is the driving algorithm, it probably also averages things out these days and does not follow every momentary stray the NN might suggest.

Think about what this means for FSD though. It needs to be accurate and consistent to navigate on urban roads, especially around car parks and the like. Remember, they promised it would come out of your garage and to your front door, find a parking space by itself etc.

Imagine it trying to navigate road works with a heavily averaged reading. It can't reliably keep in a lane as it is, probably because there is no level of averaging and fudging that both stops ping pong and avoids crossing the lines in all situations.

For FSD it has to be perfect in all situations. Right now it can't even see lines consistently, only on average.

 

[verygreen]

btw I have my own theory for why in autosteer mode on AP2 the steering wheel osculates the way it does. I suspect that's the way to detect if you are holding the wheel or not. Ever noticed that when you do hold the steering wheel firmly, you typically don't get any nags?

 

[3Victoria]

This is EAP, not FSD -- so one cannot comment on how FSD is doing.

 

[daktari]

No way they are more ahead on FSD than EAP. They must fix the basis first. The build on that with new NNs is my guess. I am using ap whenever possible, also city streets. FSD seems so far away... And Auto wipers.

 

[3Victoria]

Two (mostly) different codebases. The code will be layered, with the upper control using lower layers, such as the visual object recognition NN (VOR-NN). The VOR-NN and associated routines is substituting for the Mobileye chip in AP1. So, AP1 = Mobileye chip + driving-logic. AP2/2.5 is VOR-NN + driving-logic. FSD = VOR-NN + mapping/routing-logic + driving-logic, and I would not be surprised if the driving logic uses a second deep NN.

So the lower level share the current NN stuff, so that won't be lost in a transition to FSD.

Also, why have two product names, EAP and FSD, if they are the same but for capability? Why not just have EAP, EAP+, EAP2, EAP-MAX, EAP-Ultra?

 

[Swift]

Also, why have two product names, EAP and FSD

Because they distinguish assistance from autonomy for the purpose of charging more money of course

 

[BigD0g]

EAP == hands, FSD == no hands?

 

[croman]

EAP == hands, FSD == no hands?

I hope not. EAP should be hands free on highway. I'm not sure why it can't manage L3 with true interstate. It is starting to recognize construction and other obstacles so it stands to reason it could provide advanced notice of disconnection if the system is not robust enough to handle some common interstate challenges.