So, when is APv2 going to be noticeably better than APv1? Or is it already?

[xav-]

To whoever manages this forum, it would be nice to add a new separate section dedicated to the Autopilot.

It doesn't make sense to have many concurrent, separate Autopilot threads in the model 3, model S and Model X sections! Having everything in one place would be really helpful.

Thanks

 

[alcibiades]

To whoever manages this forum, it would be nice to add a new separate section dedicated to the Autopilot.

It doesn't make sense to have many concurrent, separate Autopilot threads in the model 3, model S and Model X sections! Having everything in one place would be really helpful.

Thanks

Autonomous Vehicles

That's where more general autonomy stuff is. But you are right, there is no Tesla Autopilot general forum.

 

[marcmerlin]

To whoever manages this forum, it would be nice to add a new separate section dedicated to the Autopilot.

It doesn't make sense to have many concurrent, separate Autopilot threads in the model 3, model S and Model X sections! Having everything in one place would be really helpful.

Thanks

You are correct. I looked for an AP forum before posting here, and did not find one.

 

[P100DHG.]

Maybe, eventually, they'll scrap the idea of homemade AP and go back to ME/Intel for AP3.0.

Doubt it. If Mobileye is worth $15 billion then anything Tesla accomplishes that is on par or beyond what Mobileye can offer makes Tesla that much more valuable. Elon has to deliver, first because of stock valuation and second because their will be lawsuits that never end.

Also the Model 3 which was originally suppose to be the EV for the people has become something more, encompassing Tesla Network ride share. 500,000 preordered! Tesla has yet to manufacture that many cars in its entire history so another reason Elon has to deliver on EAP and FSD and do it all in house. Can't start down a path of producing 500,000 cars and mid way through realize you've messed up a huge component of the car and can't deliver.

My prediction is Q4 will bring something significant and AP2 will surpass AP1 just in time for customers getting their first Model 3s.

 

[HankLloydRight]

Elon has to deliver on EAP and FSD and do it all in house.

He doesn't have to. I'm saying, if they continue down the homemade AP path, it's going to be a colossal disaster. Much less so than switching back to MobileEye. He'll never truly deliver on FSD. There's never going to be a network of driverless Model 3s roaming about to drive people around. Not in dozens of years, anyway.

I don't care how many pre-orders there are. That doesn't make it any more or less possible to create new technology.

 

[lunitiks]

HankLloyd's right

 

[HankLloydRight]

Can't start down a path of producing 500,000 cars and mid way through realize you've messed up a huge component of the car and can't deliver.

Yes, you can. It's called a sunk cost (see: Sunk cost - Wikipedia). It's better to minimize their exposure and losses and do a mid-course correction, than continuing down a path they know is wrong, just because they've already gone that far.

Tesla is already doing that with AP2.0 which they admitted can't support FSD, which they spent all of 2016 saying it absolutely would. Now all of a sudden it can't... and they're on the hook up upgrade tens-of-thousands of AP20 cars for free.

 

[P100DHG.]

He doesn't have to. I'm saying, if they continue down the homemade AP path, it's going to be a colossal disaster. Much less so than switching back to MobileEye. He'll never truly deliver on FSD. There's never going to be a network of driverless Model 3s roaming about to drive people around. Not in dozens of years, anyway.

I don't care how many pre-orders there are. That doesn't make it any more or less possible to create new technology.

You're right I used "has to", I should have said something less concrete. He doesn't have to do anything.

Tesla ending development of technology/ Tesla Vision in this case that's worth $15 billion doesn't seem likely.

 

[conman]

Tesla is already doing that with AP2.0 which they admitted can't support FSD, which they spent all of 2016 saying it absolutely would. Now all of a sudden it can't... and they're on the hook up upgrade tens-of-thousands of AP20 cars for free.

Eh, while there's circumstantial evidence of this, where have they "admitted" it?

 

[HankLloydRight]

Tesla ending development of technology/ Tesla Vision in this case that's worth $15 billion doesn't seem likely.

Who says it's worth $15 billion? Just because MobileEye sold to Intel for that? You've got to be kidding. AP2 is a far cry from the technology that MobileEye currently has under development.

By most measures, it's not even at parity with AP1 yet.

 

[HankLloydRight]

Eh, while there's circumstantial evidence of this, where have they "admitted" it?

I'm not good at searching Elon's tweets, but I think it's in there somewhere.

They've already pre-sold FSD to lots and lots of AP2.0 customers. I don't think it's going to go over very well if those owners have to cough up more cash or just swallow the fact that their cars will never FSD. It's a class-action lawsuit just waiting to happen.

 

[conman]

I'm not good at searching Elon's tweets, but I think it's in there somewhere.

And I'm quite sure it's not. He said IF a retrofit was required it would be possible or easy to do or something but did not say it was necessary. No official from Tesla has ever "admitted" that FSD was not possible with AP2.0 hardware. That's a very big distinction.

 

[HankLloydRight]

I'm certainly willing to concede that they haven't outright admitted it yet, but they did change course mid-stream to update the AP suite to add more cameras and sensors, which was my original point. If AP2.0 is fully capable of FSD, what are the extra cameras and sensors for? I think "FULL SELF DRIVING" is a pretty high bar to claim -- you can't get more autonomous than "FULL SELF DRIVING".

Even going back, Elon originally claimed AP1.0 would pull out of the garage and meet you in front of your house ("ON PRIVATE ROADS ONLY" of course).. and yeah, that's never going to happen.

 

[xav-]

I don't think the answer is intel.

I think the answer will be Lidar: A full 3D picture of the environment. That would make the code millions of times easier to write. You know exactly where you can go, the obstacles on the way etc. It's like a video game basically

Writing code on the other hand to do image recognition, to calculate distances etc from an image, is not trivial at all and is guaranteed to have false positives and false negatives

 

[Ghosty]

They should just modify all roads - incorporate a Bluetooth capable chip in every cap on the road/painted line - and let the car read off that signal. Car stopped ahead? No problem - network already advised you to slow down b/c car ahead is stopped or slowing down. \

Would work no different than circuitry network. Almost zero possibility of network error - pretty much eliminating car accidents. This would allow for FSD. But you will no longer truly be in control of your car. It is hooked to the network and you'll be transported just like data is transported.

Trademarked!

 

[12Pack]

Indeed the radar doesn't do well with soft objects and does't not deal with items outside the zone that the camera denotes as the lane occupied by the car. OTOH is has signaled impending collision for items approaching the lane at close to right angles with speed. This included a pedestrian in one case for me. (AP1)

 

[Matias]

In my experience, AP1 almost never identifies previously untracked stopped cars that you come upon..

In my experience, AP1 almost always identifies previously untracked stopped cars that you come upon.

 

[TaoJones]

In my experience, AP1 almost always identifies previously untracked stopped cars that you come upon.

First person to post dashcam video to support their assertion across, say, 5 encounters wins.

And my money's on dknisely.

 

[HankLloydRight]

Cars directly in front of you?

It most certainly does.

 

[lunitiks]

Yes, no question. If I bothered to have a dashcam I would prove it today