Coast to coast drive happening this year for all FSD Teslas!

[diplomat33]

@diplomat33 Quite the contrary, it seems problematic (or is it emblematic) to me.

Instead of working on the city driving part, Tesla is still working on perception.

Mind you I’m not saying Tesla doesn’t have any driving code as well. But you need super reliable perception to really start honing and building your driving policy which is a big problem in itself too.

I don't see why Tesla could not work on driving code using simulations at the same time that they are working on improving city street perception. Plus, Elon mentions not driving through people's driveways which is actually a separate scenario from routine city driving like driving through intersections. So city driving could be done already and Tesla is just working on these special scenarios like driving through a residential area.

 

[electronblue]

I don't see why Tesla could not work on driving code using simulations at the same time that they are working on improving city street perception. Plus, Elon mentions not driving through people's driveways which is actually a separate scenario from routine city driving like driving through intersections. So city driving could be done already and Tesla is just working on these special scenarios like driving through a residential area.

Because if you are still lacking reliable perception, you aren’t likely in a position to run and test very advanced driving policy either. Due to the limits of their perception at this time I’d expect Tesla is actually running some very simplistic city driving more reminiscent of NoA in complexity rather than something the likes of Waymo actually run.

I mean how complex driving policy can you do if you can’t even read a wide range of traffic signs reliably?

 

[diplomat33]

Because if you are still lacking reliable perception, you aren’t likely in a position to run and test very advanced driving policy either. Due to the limits of their perception at this time I’d expect Tesla is actually running some very simplistic city driving more reminiscent of NoA in complexity rather than something the likes of Waymo actually run.

I mean how complex driving policy can you do if you can’t even read a wide range of traffic signs reliably?

Uh? Who said Tesla can't read traffic signs? Elon only mentioned NN to better detect subtle curbs and driveable spaces like a person's private driveway. There was nothing in Elon's statement that would suggest that Tesla cannot read traffic signs reliably.

 

[electronblue]

Uh? Who said Tesla can't read traffic signs? Elon only mentioned NN to better detect subtle curbs and driveable spaces like a person's private driveway. There was nothing in Elon's statement that would suggest that Tesla cannot read traffic signs reliably.

From what we know, there is nothing suggesting they can read the signs reliably either (at least beyond stop signs and traffic lights). As said maybe they will pull a rabbit out of the hat and there is a secret codebase and NN hidden somewhere but so far Autopilot 2+ has been ”what you see is what you get” without any real surprises so probably better not to assume anything we haven’t seen or heard directly about...

 

[electronblue]

Frankly I would not be surprised if the initial navigate on city streets is nothing more but a map-based lane selector like NoA combined with auto steering capable of tight turns and TACC that can start and stop on traffic lights and stop signs — with the speeds still coming from a database or cloud (fleet speed) and perhaps even the traffic light locations.

Something like this could well be what Musk is testing, yet still be a million miles away from true autonomous driving that involves understanding and interpreting complex traffic rules and signs.

 

[diplomat33]

From what we know, there is nothing suggesting they can read the signs reliably either (at least beyond stop signs and traffic lights). As said maybe they will pull a rabbit out of the hat and there is a secret codebase and NN hidden somewhere but so far Autopilot 2+ has been ”what you see is what you get” without any real surprises so probably better not to assume anything we haven’t seen or heard directly about...

Frankly I would not be surprised if the initial navigate on city streets is nothing more but a map-based lane selector like NoA combined with auto steering capable of tight turns and TACC that can start and stop on traffic lights and stop signs — with the speeds still coming from a database or cloud (fleet speed) and perhaps even the traffic light locations.

Something like this could well be what Musk is testing, yet still be a million miles away from true autonomous driving that involves understanding and interpreting complex traffic rules and signs.

You could be right, maybe. We shall see when Tesla releases "automatic city driving" and the new AP3 software. But keep in mind that Karpathy did say back in Oct 2018 that Tesla had much larger NN that only work on AP3. And we also know that "automatic city driving" requires AP3. So we can infer that Tesla probably does have, as you put it, some "secret codebase and NN hidden somewhere" that we have not seen yet. After all, isn't that kind of the whole point of AP3? You seem to be making the assumption that there is no secret AP3 software, that the current AP2 software is as good as it gets. So you have come up with an implementation of NOA on city streets that would work with the current AP2 software. But if that's the case, why bother with AP3? Why not release NOA on city streets now on AP2?

Honestly, I don't think we really know for sure yet.

 

[electronblue]

@diplomat33

Two points

1. We have been in the ”secret codebase” territory before. Not so long ago there was a widespread belief that Tesla had a secret FSD codebase separate from the early (E)AP2 codebase. Of course this pipedream didn’t pan out. Excuse my scepticism this time around, then.

2. What Musk and Karpathy actually have been saying about the specific features they are testing right now does not instill confidence in me that they’d be much more advanced than what I outlined above.

The examples Musk has been mentioning about FSD has been traffic lights, stop signs, roundabouts I believe, that type of thing. And now curbs etc.

If they had basic perception figured out he’d be talking unprotected left turns without safety drivers and complex right of way incidents with a bike partially blocking the way or somesuch.

Instead we’re still talking curbs and stop signs and traffic lights. It doesn’t even sound like they have a full range of obstacle detection figured out (unusual objects for example).

I would say history is on my side on this one and yours is more wishful thinking. Now history can of course be wrong but it seems like a leap of faith more than actual data.

 

[Phrixotrichus]

Honestly, I don't think we really know for sure yet.

I`m seriously astonished that after so many broken promises some fans still give Elon the benefit of the doubt.....

I´ll not believe a thing Tesla claims or Elon promises anymore if a corresponding product isn`t actually available.

 

[electronblue]

I`m seriously astonished that after so many broken promises some fans still give Elon the benefit of the doubt.....

I´ll not believe a thing Tesla claims or Elon promises anymore if a corresponding product isn`t actually available.

By his own admission @diplomat33 hasn’t lived through all those broken promises and thus — no matter what he says — can’t quite relate to them and take them that seriously. He genuinely seems to believe things will be different now even though personally I can’t see the signs regarding Musk are good at all compared to what they used to be when we earlier customers gave him the benefit of the doubt... at least there seemed to be better reason to believe Musk back then and it was still wrong to do so...

 

[diplomat33]

By his own admission @diplomat33 hasn’t lived through all those broken promises and thus — no matter what he says — can’t quite relate to them and take them that seriously. He genuinely seems to believe things will be different now even though personally I can’t see the signs regarding Musk are good at all compared to what they used to be when we earlier customers gave him the benefit of the doubt... at least there seemed to be better reason to believe Musk back then and it was still wrong to do so...

No, I was not an owner through all the broken promises of the past. So I am not jaded. I am hopeful based on what I see that things will work out, yes.

@diplomat33
1. We have been in the ”secret codebase” territory before. Not so long ago there was a widespread belief that Tesla had a secret FSD codebase separate from the early (E)AP2 codebase. Of course this pipedream didn’t pan out. Excuse my scepticism this time around, then.

Big difference. The theory that Tesla had a separate codebase for FSD and EAP was nothing more than a fan theory. Karpathy actually confirmed that the AP3 codebase exists.

 

[boonedocks]

By his own admission @diplomat33 hasn’t lived through all those broken promises and thus — no matter what he says — can’t quite relate to them and take them that seriously. He genuinely seems to believe things will be different now even though personally I can’t see the signs regarding Musk are good at all compared to what they used to be when we earlier customers gave him the benefit of the doubt... at least there seemed to be better reason to believe Musk back then and it was still wrong to do so...

Well said sir. I do like @Diplomat33’s optimism.....but.....the truth is reality is on our side. I too once took every word Elon said in good faith ....as those of us that have been around a while know Elon is full of good intentions. Good intentions however don’t make reality.

I have learned to accept Tesla cars for what they offer today (on our 4th as of last Sunday) and am hopeful for the future but certainly not intoxicated by it and Elon anymore.

 

[diplomat33]

Well said sir. I do like @Diplomat33’s optimism.....but.....the truth is reality is on our side. I too once took every word Elon said in good faith ....as those of us that have been around a while know Elon is full of good intentions. Good intentions however don’t make reality.

I have learned to accept Tesla cars for what they offer today (on our 4th as of last Sunday) and am hopeful for the future but certainly not intoxicated by it and Elon anymore.

I am optimistic about Tesla's FSD but I do not believe Elon's timeline that Tesla will have robotaxis in 2020.

 

[SVMike]

The NN is running regardless. The additional neurons for rain sense are not an appreciable amount of energy.

This is not true, that we know of. Unless there is ONE NN model that does everything.
The new NN chip is running anyway, but if it is 20x faster than the older chips, it will likely do its task(s) then be idle for some time, then get some more work will get scheduled later...such as every 26 ms or so. (the Tesla Cam videos have an odd frame rate of 37.37 fps). The power consumption will be less with more idle time.

 

[SVMike]

Considering how long the NN rain sensing was missing and how poor the NN rain sensing is still... with first AP2 cars are reaching the end of natural ”new car” lives this year (lease periods etc) it is clear they would have been better served by a rain sensor. It would also have given Tesla some redundancy on rain sensing down the road in tricky lighting situations.

If Tesla indeed was skating to where the puck is (I’m not sure if it wasn’t simply misguided cost-savings hubris from Elon’s part), they failed.

I submitted another auto wiper bug the other day.
Perfectly clear and dry day driving home in the evening to the west.
Glaring sun at a stop light with the sun partially blocked by a 5 or 6 story building.
While stopped, suddenly the wipers came on at max wiping test for 6 or 8 wipes.
I finally brought up the wiper card and was going to stop them...of course there is no stop.
I tapped on the slow speed then back to auto and got them to go off

 

[mongo]

This is not true, that we know of. Unless there is ONE NN model that does everything.
The new NN chip is running anyway, but if it is 20x faster than the older chips, it will likely do its task(s) then be idle for some time, then get some more work will get scheduled later...such as every 26 ms or so. (the Tesla Cam videos have an odd frame rate of 37.37 fps). The power consumption will be less with more idle time.

Sure, on HW3 with overhead, there will be idle time, which the HW3 processor will be likely be great at taking advantage of.
Given the power needed for motion, HVAC, lighting, and FSD NN, a rain sensing NN (which I feel is needed as part of occlusion detection) does not meet the claim of

Aside from the fact that it barely works, using a NN wastes a lot of energy.

which is what I responding to. The whole HW3 computer peaks at less than 100W, and the rain NN is likely << smaller that the FSD/ AP net. Esp since wiper rain sense only runs on 3 cameras at most.

 

[Eno Deb]

I submitted another auto wiper bug the other day.
Perfectly clear and dry day driving home in the evening to the west.
Glaring sun at a stop light with the sun partially blocked by a 5 or 6 story building.
While stopped, suddenly the wipers came on at max wiping test for 6 or 8 wipes.

The same thing happened to me today while I was in a covered parking structure. Really weird. Never had this before. Perhaps a bug in the 2019.20.1 firmware.

I finally brought up the wiper card and was going to stop them...of course there is no stop.
I tapped on the slow speed then back to auto and got them to go off

There is a stop actually. Just tap the wiper icon above the speed/auto selection bar. It turns off the wiper altogether.

 

[conman]

There is a stop actually. Just tap the wiper icon above the speed/auto selection bar. It turns off the wiper altogether.

That sounds like a model 3 thing...

 

[DrDabbles]

@mongo I for one do get the point why NN weather detection can be useful. But I do think on this one Tesla would have been better off leaving the rain sensor out only after the NNs work well.

/thread right here.

 

[Bladerskb]

I don't see why Tesla could not work on driving code using simulations at the same time that they are working on improving city street perception. Plus, Elon mentions not driving through people's driveways which is actually a separate scenario from routine city driving like driving through intersections. So city driving could be done already and Tesla is just working on these special scenarios like driving through a residential area.

These are not special scenarios. These are building blocks of a true self driving system.
As I outlined, it shows how behind tesla is in making an actual self driving system.

"Road edge and curbs are two neural networks we are developing right now" - Elon Musk

In contrast Mobileye has this in production since Q4 2017.

Shows you they still have alot of catching up to do.
All the networks below they need and they are just now developing them as Elon said and this confirms the fact that these specific networks weren't mentioned in autonomy day

https://pbs.twimg.com/media/D6aOEGRXoAEsKG7.jpg:large

 

[diplomat33]

These are not special scenarios. These are building blocks of a true self driving system.
As I outlined, it shows how behind tesla is in making an actual self driving system.

"Road edge and curbs are two neural networks we are developing right now" - Elon Musk

In contrast Mobileye has this in production since Q4 2017.

Shows you they still have alot of catching up to do.
All the networks below they need and they are just now developing them as Elon said and this confirms the fact that these specific networks weren't mentioned in autonomy day

https://pbs.twimg.com/media/D6aOEGRXoAEsKG7.jpg:large

Yes, Tesla has a lot of catching up to do. That is not news to me. But I trust Tesla will get there and when they do, they can upload it to hundreds of thousands of cars already on public roads.