Why AP 2.0 Won't Be Here Soon, and It Won't Be What You Think It Is

[MarkS22]

I agree with glare as well. Again, having the cameras six feet apart would help there. I agree glare is a more likely example.

But I mean if it's not a leaf, what about bird crap? The wiper could smear and make it worse. Rare? Perhaps. But corner cases are what causes problems.

 

[Alketi]

An unanswered question (wish someone had asked on the CC) is whether or not AP 2.0, as currently designed, has enough redundancy to meet regulator approval for true driverless operation (L5). Granted these decisions are at least a year or two away, but it's an important future-proofing question for such a large purchase.

This would potentially imply that beyond redundant forward cameras, things like redundant ABS braking, power supplies, and processors are also needed. Some of this is discussed in an Infineon presentation here (page 10): http://www.soiconsortium.org/fully-depleted-soi/presentations/silicon-saxony-day-2015/1. SOI1_Infineon_Adlkofer.pdf

 

[Canuck]

An unanswered question (wish someone had asked on the CC) is whether or not AP 2.0, as currently designed, has enough redundancy to meet regulator approval for true driverless operation (L5). Granted these decisions are at least a year or two away, but it's an important future-proofing question for such a large purchase.

That's a good question. Elon Musk even said one year ago:

"For full autonomy you’d obviously need 360 cameras, you’d probably need redundant forward cameras, you’d need redundant computer hardware, and like redundant motors and steering rack. For full autonomy you’d really want to have a more comprehensive sensor suite computer systems that are fail proof."

So I see missing from AP2.0 redundant computer hardware, motors and steering rack. But that's just Elon's opinion and we know it's subject to change. The question you raise has no answer at this time.

 

[mikes_fsd]

So I see missing from AP2.0 redundant computer hardware, motors and steering rack.

All, Dual Motor Tesla's have redundant motors. It would not be optimal for efficiency, but it would work in case of one motor failure.

The 3 forward facing cameras are a form of redundancy.
So, really what you are missing is the computer hardware and steering rack - right now it is probably not redundant only because of cost overhead given they are not done with <L5 software yet.

 

[LargeHamCollider]

I don't think redundancy is missing in the computer hardware, we don't know if it is (or at least I don't), but we do know that nVidia advertised that two PX2.0 boards are needed for full autonomy, I would bet that AP2.0 has two PX2.0 boards.

 

[MarkS22]

The fact is, they're taking a gamble. It's almost like buying a router with a new WiFi standard that hasn't been finalized.

Like those manufactures, it's a calculated gamble where they think firmware can address the concerns of regulators and interoperability.

I think another thing oddly absent (but maybe just not announced) is "Car to Car" and "Car to Infrastructure" communication. Many people think that's a necessity for L5 (including former allies, MobilEye). Of course, Elon noted the computer is upgradable but I'd hope they don't have to retrofit new WiFi systems. Or, as others have suggested, perhaps $1,000 of the cost is to allow for one future hardware upgrade?

 

[Saghost]

I think another thing oddly absent (but maybe just not announced) is "Car to Car" and "Car to Infrastructure" communication. Many people think that's a necessity for L5 (including former allies, MobilEye).

I don't think the standards for that exist yet, which makes it hard for Tesla to comply with them. On the other hand, every S and X already speaks on pretty much all the frequency bands allowed for a mobile device - cellular, Wi-Fi, Bluetooth - so it should just be a firmware update once a standard is agreed upon, and having all these Autonomous-Ready cars driving around should give Tesla some leverage as folks are deciding the standard to make it something they can just dump into new firmware.

 

[scottg96]

My biggest concern is putting all three forward cameras in a single location. That's the failure point. A single wet leaf can knock it out. A human has the ability to lean around windshield obstructions.

I would've expected redundant locations (like top-left and top-right corners of the windshield) for both safety, better imaging of the forward view, and seeing around forward obstacles.

Even if the radar is enough to pull you over, it doesn't seem optimal.

I think I recall hearing Elon mention during the Q&A call that the location of the cameras is within the range of the windshield wipers - I imagine those would be able to remove obstructions like that, otherwise it would let you know it's severely impaired and needs to pull over, for example.

 

[scottg96]

Sorry, I didn't bother reading the rest of the thread where someone else brought that up and you addressed it - just joined TMC.

 

[calisnow]

My biggest concern is putting all three forward cameras in a single location. That's the failure point. A single wet leaf can knock it out. A human has the ability to lean around windshield obstructions.

You question is a good one - and also obvious. Given that it's obvious let's assume Tesla thought of that. The question becomes - why did they choose to do what they did?

The fact is, they're taking a gamble. It's almost like buying a router with a new WiFi standard that hasn't been finalized...I think another thing oddly absent (but maybe just not announced) is "Car to Car" and "Car to Infrastructure" communication. Many people think that's a necessity for L5 (including former allies, MobilEye). Of course, Elon noted the computer is upgradable but I'd hope they don't have to retrofit new WiFi systems. Or, as others have suggested, perhaps $1,000 of the cost is to allow for one future hardware upgrade?

Tesla's entire product has been a gamble for the last - what - 10 years? Yet we buy their cars anyway. As for car-to-car communications - yep, many people think it's necessary - such as MBZ with their excellent lane keeping (oh wait...). Only Elon Musk is actually building product. If I waited around for the auto industry to agree on a bunch of standards I'd be old and gray. If things are different 5 years from now - buy again. There's no other solution for anyone wanting to be on the bleeding edge of self driving automobiles.

 

[calisnow]

Two Mercedes talking to each other via MBZ's new car-to-x standard for the 2017 E-class:

MBZ 1: "Hey fellow car 1000 feet behind me - just wanted you to know there's a road here where I am. I can't find the lanes, but I know they're around here somewhere."

MBZ 2: "Roger that, good to know."

 

[calisnow]

BTW @MarkS22 - I don't have any answer to your question either. I have no idea why they put the three camera cluster in the same location - it does seem to be, at first glance anyway, a vulnerability.

 

[MarkS22]

You question is a good one - and also obvious. Given that it's obvious let's assume Tesla thought of that. The question becomes - why did they choose to do what they did?



Tesla's entire product has been a gamble for the last - what - 10 years? Yet we buy their cars anyway. As for car-to-car communications - yep, many people think it's necessary - such as MBZ with their excellent lane keeping (oh wait...). Only Elon Musk is actually building product. If I waited around for the auto industry to agree on a bunch of standards I'd be old and gray. If things are different 5 years from now - buy again. There's no other solution for anyone wanting to be on the bleeding edge of self driving automobiles.

I think it's more a gamble on regulation. For example, if it ends up requiring specific communications or specific sensor requirements.

 

[MarkS22]

Two Mercedes talking to each other via MBZ's new car-to-x standard for the 2017 E-class:

MBZ 1: "Hey fellow car 1000 feet behind me - just wanted you to know there's a road here where I am. I can't find the lanes, but I know they're around here somewhere."

MBZ 2: "Roger that, good to know."

Hah. I'm just envisioning a line of MBZs looking like the game Centipede, going over a cliff.

 

[Reeler]

Wouldn't the MBZ run over pedestrians before killing the drivers by running off a cliff?

 

[techmaven]

The fact is, they're taking a gamble. It's almost like buying a router with a new WiFi standard that hasn't been finalized.

Like those manufactures, it's a calculated gamble where they think firmware can address the concerns of regulators and interoperability.

I think another thing oddly absent (but maybe just not announced) is "Car to Car" and "Car to Infrastructure" communication. Many people think that's a necessity for L5 (including former allies, MobilEye). Of course, Elon noted the computer is upgradable but I'd hope they don't have to retrofit new WiFi systems. Or, as others have suggested, perhaps $1,000 of the cost is to allow for one future hardware upgrade?

Car to car, and car to infrastructure communications are clearly not required for L5. Further, if you do have to rely on them, it's a clear case where your effort is going to fail. Any team that talks about V2V or V2I as part of their ADAS SAE Level 5 effort is clearly not going to ship anytime in the next decade. All you have to do is think about it a little bit... it's almost impossible for a vehicle to depend on V2V or V2I as part of their driving. Think of all the failure modes that would have to be handled. And if you can't depend on it, it means you can't use it at all. It's absurd to be talking about either in a phase where manual driving and automatic driving share the road. One day, if we get to a point where manual driving is outlawed on public roads, then you can replace some parts of street signs, traffic signals and the like with V2I. As for V2V, that's even more absurd. Unless what you mean is the same kind of traffic notification that we get now with crowd sourcing congestion.

Listen to this from George Hotz:

He may be very wrong in some things, but if you think about what he says about V2V and V2I, it makes sense.

 

[calisnow]

@techmaven - I think there is a bug in the forum software. I hit the quote button on your post and what came out was another paragraph you must have written after you ended your sentence with "...absurd."

I'm not going to post a screenshot here publicly because you obviously deleted what you wrote (it was nothing remotely salacious or anything folks - just he must have finished a thought he edited out). But I am sending you PM with the screenshot so you can see that this happened.

This is concerning - you can type something, then edit it - and yet the quote function still pulls up the deleted text.

 

[calisnow]

Now wait one second - this time I see what you added actually on your public post - the George Hotz video. The first time I did not see it publicly - maybe you were just then typing it in? Anyway I sent you the screenshot - you can see the disparity between the publicly visible text at that time and what the quote function pulled up.

 

[techmaven]

Yeah, I added some stuff to the comment... including the video.

 

[MarkS22]

Car to car, and car to infrastructure communications are clearly not required for L5. Further, if you do have to rely on them, it's a clear case where your effort is going to fail. Any team that talks about V2V or V2I as part of their ADAS SAE Level 5 effort is clearly not going to ship anytime in the next decade. All you have to do is think about it a little bit... it's almost impossible for a vehicle to depend on V2V or V2I as part of their driving. Think of all the failure modes that would have to be handled. And if you can't depend on it, it means you can't use it at all. It's absurd to be talking about either in a phase where manual driving and automatic driving share the road. One day, if we get to a point where manual driving is outlawed on public roads, then you can replace some parts of street signs, traffic signals and the like with V2I. As for V2V, that's even more absurd. Unless what you mean is the same kind of traffic notification that we get now with crowd sourcing congestion.

Listen to this from George Hotz:

He may be very wrong in some things, but if you think about what he says about V2V and V2I, it makes sense.

That doesn't mean regulators won't require it to ensure it's eventually enabled on all vehicles at a base level. I agree it won't technically be necessary, but the more cars that have it, the safer (or at least more efficient for traffic) it's likely to be.