Poll: Testing the wisdom of crowds - how far off is Autopilot V2 Hardware?

[Alketi]

There was at least one Tesla spotted with prototype camera/sensor hardware on the front end, so they're definitely testing another revision.

A couple of questions calisnow:

1. Are you guessing that the next hardware revision will be capable of supporting (though not actually delivering) Level 4 autonomy? [This would be sooner than the more powerful EyeQ4 chip arrives, but would fit the definition of a "big" end-of-year announcement from Elon]

2. Would you expect the Model 3s rolling off the assembly line have the same sensor suite? [I would expect, if the hardware costs are reasonable, that the 3 would have the same capability to minimize the software development/support tasks]

 

[TaoJones]

About 5 months ago, Elon said we'd get traffic light and stop sign features in about a year. Nobody followed up to ask if AP 1.0 (current cars up until the recent tri-camera) would be sufficient to deliver those new features.

Obviously AP 2.0 will, and probably the expected EOY front camera update will.

As much as I'd like traffic light and stop sign reaction (different from recognition) with AP 1.0, as time passes I dunno.

May have been a Muskism with an unspoken "but we've concluded it would be safer if you bought a new car". I can't remember the context of the question.

Perhaps the current tri-camera or the EOY cams will be retrofittable, but...

 

[Skione65]

Well I have one autopilot S already - I need a second one, so I'm waiting. You're right about what Elon has said, but based on what he did last time, many of us believe that what is going to happen is this:

1 - By December 2016 the hardware will be updated to have at least three cameras facing the front
2 - The fleet of AP 2.0 hardware Teslas will drive around for 18 more months collecting data, but without the full autonomy turned on.
3 - Steps 1 & 2 put you at the 24 month mark, and then you have the "additional year" Elon mentioned for the cars to demonstrate they are safe and for regulators to allow people to legally drive while not paying attention.

The evidence for a very imminent hardware update is overwhelming at this point - I can go into more detail on it and list it off if you want to hear it.

@calisnow

Yes.....I'm very interested in the details. I as well as others would greatly appreciate you taking the time out of your busy days to give us further details.

Looking forward to it!

Ski

 

[Zybd1201]

Well I have one autopilot S already - I need a second one, so I'm waiting. You're right about what Elon has said, but based on what he did last time, many of us believe that what is going to happen is this:

1 - By December 2016 the hardware will be updated to have at least three cameras facing the front
2 - The fleet of AP 2.0 hardware Teslas will drive around for 18 more months collecting data, but without the full autonomy turned on.
3 - Steps 1 & 2 put you at the 24 month mark, and then you have the "additional year" Elon mentioned for the cars to demonstrate they are safe and for regulators to allow people to legally drive while not paying attention.

The evidence for a very imminent hardware update is overwhelming at this point - I can go into more detail on it and list it off if you want to hear it.

I'd love to hear more about this too.

I am not quite sure how you can make such definitive predictions. The past informs the future but does not predict it.

It is entirely possible that full autonomy can be achieved with the hardware being built into the Model S as of June, 2016.

A trifocal camera is not necessarily required to obtain full autonomy.

 

[IngTH]

I have question related to the front radar.

Do you guys know which brand/model is the front LRR in a current model S/X?

 

[MarkS22]

I have question related to the front radar.

Do you guys know which brand/model is the front LRR in a current model S/X?

Bosch. Not sure the exact model but it's 3.9mm wavelength, per Elon's tweets.

Likely: http://www.bosch-presse.de/presseforum/pressdownload/text/Vorschau I Preview IAA 2013.pdf?id=6395,2

 

[Saghost]

I'd love to hear more about this too.

I am not quite sure how you can make such definitive predictions. The past informs the future but does not predict it.

It is entirely possible that full autonomy can be achieved with the hardware being built into the Model S as of June, 2016.

A trifocal camera is not necessarily required to obtain full autonomy.

Neither the front camera nor the front radar can see far enough to the sides to monitor cross traffic, so I don't see how you're expecting the current package to handle stop signs or turning on red at traffic lights.

I think it's safe to assume at least the triple front camera is required for full autonomy (and the current rear camera hooked in to AP, too.) Of course, those two are probably retrofittable.

My guess is they'll go with more cameras than that and 360 coverage. Cameras are cheap, and the processing power to handle them not much more expensive.

 

[MarkS22]

I've gotta say, all this talk of OTA software updates, 8.0 being the biggest update since 1.0, talking with Bosch and Mobileye about current hardware, upgrading the radar software to do point clouds... makes me think 2.0 hardware suite is even further out. That's a lot of resources going into development for hardware that'll be replaced "soon."

Not to say I think the current hardware is capable of full autonomy. It's not. As many have noted, it needs a much wider FOV (especially for cross-traffic and right-angle turns). Based on this week's comments, I think it's more likely they'll try to take this hardware as far as it can for highway driving. Potentially some level of on-ramp to off-ramp "autonomy." Once that's sufficiently improved and hundreds of millions of miles of data collected, then they'd use this to develop the 2.0 sensor suite.

I could be wrong, but Tesla is in an incredibly unique position. By watching all this data, they're in a much better position to figure out exactly what sensors are needed in the real world. And they might need more or less than current experts and simulations are predicting. Absolutely nobody will know the corner cases and trouble spots better than Tesla in real world settings over hundreds of millions of miles.