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

[techmaven]

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.

How does it make it safer? A SAE Level 5 car has to work as the only car within a flotilla of manual driving cars and in situations where there is zero V2I infrastructure. It has to do this with utmost safety. How much safer would providing either of these make things? Instead, if there are problems, there are likely problems for manual drivers already... and that's a normal part of traffic engineering to solve those issues.

 

[MarkS22]

How does it make it safer? A SAE Level 5 car has to work as the only car within a flotilla of manual driving cars and in situations where there is zero V2I infrastructure. It has to do this with utmost safety. How much safer would providing either of these make things? Instead, if there are problems, there are likely problems for manual drivers already... and that's a normal part of traffic engineering to solve those issues.

Make no mistake, without communication, it's going to be safer than humans. Actuaries are predicting ~80% fewer accidents but not perfection. Even Tesla is saying twice as good as a regular driver.

How much safer can it get? It'll be incremental, but we shouldn't dismiss any safety technology. It'll be particularly useful during the transition when manual cars are sharing telemetry.

In a study with 3,000 equipped cars, NHTSA concluded car to car alone could save 1,000 lives a year and prevent 500,000 accidents. The Department of Transportation is already drafting rules that could make it mandatory. Would AP2 have eliminated every single one of those? Who knows? But the fact is, it works and works well.

I've seen examples where there's an accident and cars five or six back are notified and slow before it's seen. That's a significant safety feature. (It's not unlike the much touted radar seeing two cars ahead.) The more cars that have it, the better it'll work.

There are plenty of valid reasons to want a base standard of car and infrastructure communication in Level 5 regulations. Why wouldn't you want it?

 

[techmaven]

Make no mistake, without communication, it's going to be safer than humans. Actuaries are predicting ~80% fewer accidents but not perfection. Even Tesla is saying twice as good as a regular driver.

How much safer can it get? It'll be incremental, but we shouldn't dismiss any safety technology. It'll be particularly useful during the transition when manual cars are sharing telemetry.

In a study with 3,000 equipped cars, NHTSA concluded car to car alone could save 1,000 lives a year and prevent 500,000 accidents. The Department of Transportation is already drafting rules that could make it mandatory. Would AP2 have eliminated every single one of those? Who knows? But the fact is, it works and works well.

I've seen examples where there's an accident and cars five or six back are notified and slow before it's seen. That's a significant safety feature. (It's not unlike the much touted radar seeing two cars ahead.) The more cars that have it, the better it'll work.

There are plenty of valid reasons to want a base standard of car and infrastructure communication in Level 5 regulations. Why wouldn't you want it?

Because it causes an increase in complexity for little gain. And when you start going down the failure modes, lots more corner cases crop up. Simply, the overall behavior of autonomous driving systems makes things behave better. Do you think schools of fish need fish to fish communications like those that are proposed in order to behave like they are all coordinated?

 

[MarkS22]

Because it causes an increase in complexity for little gain. And when you start going down the failure modes, lots more corner cases crop up. Simply, the overall behavior of autonomous driving systems makes things behave better. Do you think schools of fish need fish to fish communications like those that are proposed in order to behave like they are all coordinated?

We don't know exactly how fish school, but it's not a comparable system. There's research that they only focus on a single closest fish near them. This is more akin to highway driving where everyone has the same goal. As we see with AP 1.0, this is the easiest to do. The benefit of non-visual communication comes into play when you have a variety of goals where you're going in a different directions. For example, when one school is approaching another species or a portion decides to break away. That's why it's incredibly more difficult once you leave that off ramp.

Tesla is already using car to car communication, only not in realtime. That's the whole point of fleet learning. If you can move that fleet learning to realtime for a local area, many believe there would be significant safety gains.

But I digress. The point isn't the benefits of car communication. The Department of Transportation has shown interest in making it a requirement. Per your argument, it might be deemed too expensive or complicated for limited safety or traffic benefits. But many are pursuing it. I was just giving a hypothetical example of where a sensor or wireless system might be required before a government body signs off on level 5.

 

[EarlyAdopter]

You could achieve 2x safety with full L5 autonomy even if the vehicle drove no better than a human simply for the fact that the occupant could sit in the middle seat of the second row.

 

[techmaven]

We don't know exactly how fish school, but it's not a comparable system. There's research that they only focus on a single closest fish near them. This is more akin to highway driving where everyone has the same goal. As we see with AP 1.0, this is the easiest to do. The benefit of non-visual communication comes into play when you have a variety of goals where you're going in a different directions. For example, when one school is approaching another species or a portion decides to break away. That's why it's incredibly more difficult once you leave that off ramp.

Tesla is already using car to car communication, only not in realtime. That's the whole point of fleet learning. If you can move that fleet learning to realtime for a local area, many believe there would be significant safety gains.

But I digress. The point isn't the benefits of car communication. The Department of Transportation has shown interest in making it a requirement. Per your argument, it might be deemed too expensive or complicated for limited safety or traffic benefits. But many are pursuing it. I was just giving a hypothetical example of where a sensor or wireless system might be required before a government body signs off on level 5.

I am sorry if I came across as gruff over this. I have spent some time looking as various critics as well as some of the research. The amount of skepticism is outsized, partially because Tesla is not doing things the way many other efforts seem to have gone. What is fascinating to me is how some very smart people have gone down avenues of development that really do not look like directions that will actually be able to ship anything. V2V and V2I look both like things that, if your ADAS program relied on them in any way, would actually prevent your solution from shipping. After we have full SAE level 5 autonomy, then we likely can talk about some of these.

For example, I have been thinking of when we no longer allow manual driving, do we need trafffic lights at all? Do we even stop at intersections? With V2I, one locks into a timing pulse and enters virtual queues at a particular speed... and interleave through an intersection of trafffic flow allows it. But not a single car can be manually driven in this case. And you still have to have full mitigation for additional failure modes if the timing signal is either missing or erratic. Imagine the horrifying result if someone hacked the timing signal.

 

[MorrisonHiker]

I am sorry if I came across as gruff over this. I have spent some time looking as various critics as well as some of the research. The amount of skepticism is outsized, partially because Tesla is not doing things the way many other efforts seem to have gone. What is fascinating to me is how some very smart people have gone down avenues of development that really do not look like directions that will actually be able to ship anything. V2V and V2I look both like things that, if your ADAS program relied on them in any way, would actually prevent your solution from shipping. After we have full SAE level 5 autonomy, then we likely can talk about some of these.

For example, I have been thinking of when we no longer allow manual driving, do we need trafffic lights at all? Do we even stop at intersections? With V2I, one locks into a timing pulse and enters virtual queues at a particular speed... and interleave through an intersection of trafffic flow allows it. But not a single car can be manually driven in this case. And you still have to have full mitigation for additional failure modes if the timing signal is either missing or erratic. Imagine the horrifying result if someone hacked the timing signal.

Rush hour in an autonomous intersection:

A more serious look:

 

[henderrj]

I finally had time to sit and carefully watch the video again. There was something that triggered an old memory. I think I might be right – the B&W camera views didn’t look quite right. So I slowed the video to ¼ speed. I think I’m seeing a bit of infrared in them. Very difficult to tell but I was particularly struck by the brightness of the two red cars (the Fiat on the highway and the parked Tesla). Seemed too bright for simply black and white. Also the pedestrian’s skin tones. A few decades ago when I was taking some portraits we used to use a black and white film that had some infrared sensitivity in it for people with complexion problems – smoothed it out a little. It feels like I’m seeing that with these cameras. The leaves and more just seemed like there was some infrared being transferred to the visual spectrum. But very difficult to tell with sunlight.

If the camera sensitivity goes beyond the visible into infrared then the system will have better vision at night, in the fog, etc. Maybe lidar isn’t necessary. Maybe this hardware will suffice for even level 5.

Maybe.

 

[Saghost]

I finally had time to sit and carefully watch the video again. There was something that triggered an old memory. I think I might be right – the B&W camera views didn’t look quite right. So I slowed the video to ¼ speed. I think I’m seeing a bit of infrared in them. Very difficult to tell but I was particularly struck by the brightness of the two red cars (the Fiat on the highway and the parked Tesla). Seemed too bright for simply black and white. Also the pedestrian’s skin tones. A few decades ago when I was taking some portraits we used to use a black and white film that had some infrared sensitivity in it for people with complexion problems – smoothed it out a little. It feels like I’m seeing that with these cameras. The leaves and more just seemed like there was some infrared being transferred to the visual spectrum. But very difficult to tell with sunlight.

If the camera sensitivity goes beyond the visible into infrared then the system will have better vision at night, in the fog, etc. Maybe lidar isn’t necessary. Maybe this hardware will suffice for even level 5.

Maybe.

I wouldn't have been surprised to find out that the cameras were sensitive to near infrared - digital cameras are by nature, unless a filter has been installed to block it.

However, I this case I think it is clear that a filter is blocking infrared. The foliage is a reliable indicator - all living plants glow brightly in near infrared, because of the chlorophyll. In the video, the leaves don't show up white like they should with IR.

 

[MarkS22]

I am sorry if I came across as gruff over this. I have spent some time looking as various critics as well as some of the research. The amount of skepticism is outsized, partially because Tesla is not doing things the way many other efforts seem to have gone. What is fascinating to me is how some very smart people have gone down avenues of development that really do not look like directions that will actually be able to ship anything. V2V and V2I look both like things that, if your ADAS program relied on them in any way, would actually prevent your solution from shipping. After we have full SAE level 5 autonomy, then we likely can talk about some of these.

I always welcome counterpoints. I see C2C and C2X (or whatever acronym they end up using) as incremental. Almost like having radar on an airplane while flying VFR. In other words, totally capable without it, but a "nice to have" that will eventually make things safer and more efficient in those corner cases.

I was just theorizing that, to jump start mass adoption, I can see regulators say "Okay, you want a license for Level 5, then you need C2C and C2X using 802.11p." It seems like as good a starting point as any.

But I do appreciate and see your points.

 

[S4WRXTTCS]

You will probably not see AP 2.0 hardware for another few quarters at the very earliest. And that's being optimistic.

OOPS.

I'm glad I'm not Todd. That was on Sep 12th that he made the prediction.

Just goes to show you can't predict anything with Tesla.

 

[Todd Burch]

OOPS.

I'm glad I'm not Todd. That was on Sep 12th that he made the prediction.

Just goes to show you can't predict anything with Tesla.

I may have been off a few quarters on the hardware prediction (and in fact, I'm still not confident the hardware Tesla is now installing allows true L5 autonomy to the point where you could sleep in the back seat), but I don't think the overall gist of the prediction is off by much.

 

[tonglaji]

I may have been off a few quarters on the hardware prediction (and in fact, I'm still not confident the hardware Tesla is now installing allows true L5 autonomy to the point where you could sleep in the back seat), but I don't think the overall gist of the prediction is off by much.

Are you a lawyer?

 

[Todd Burch]

Are you a lawyer?

Engineer, and a realist. The focus of my original post was about when L5 SOFTWARE will be ready, and when it will be legal for your car to drive to you with no occupant. Yes, I didn't expect that the hardware would be ready quite this early, but really that wasn't the focus of the OP. (And Tesla is taking a risk that the hardware is sufficient, when in fact it may not be).

It still stands that the software is not ready, and I still believe level 5 software will not be ready for years. Tesla may have software that will drive from NY to LA by the end of next year in a best-case scenario, but I still maintain that it won't handle a lot of scenarios, and it won't be legal to do without an occupant for quite awhile after that.

If you disagree, well...then I think you're overly optimistic.

 

[⚡️ELECTROMAN⚡️]

When making predictions like this you should be prepared to be wrong. It was a well written case for why we wouldn't be seeing new AP hardware this year. It's here. It's totally okay that you were wrong. I will also be surprised if they reach full autonomy with this hardware, but the fact that they are trying is awesome. It should take us well beyond AP1 abilities and safety..

 

[Canuck]

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

It's here and it much, much more than anyone thought it would be -- the announcement is anyways and that's all we can go by at this stage.

I may have been off a few quarters on the hardware prediction (and in fact, I'm still not confident the hardware Tesla is now installing allows true L5 autonomy to the point where you could sleep in the back seat), but I don't think the overall gist of the prediction is off by much.

No it's not off by much. It's off by eons and eons.

But good for you for sticking your neck out and making predictions. So what if you're wrong? It's no big deal being wrong (unless your last name is Trump). You made good arguments for your point of view and this thread has 475 replies and counting of some very interesting content. That's nothing to be ashamed of starting.

 

[bob_p]

Though some of those who are disappointed because they purchased an AP 1.0 in September may have made their decision based on this thread, believing that AP 2.0 was nowhere close to being announced.

 

[u00mem9]

Can we wait until it's here before we start claiming it's here? It's like a Tesla release event in here!

 

[Canuck]

Can we wait until it's here before we start claiming it's here? It's like a Tesla release event in here!

Spoken by someone who didn't miss out on AP1.0. It's no big deal because it took a year to get it working right? Well I got my car shortly before without it and I can tell you it was "here" when it came out, regardless of the time it took to get it working. The line was drawn on the vehicles with the release of the hardware - not the software.

Much more hardware and a much faster processor are also here -- right NOW. They are on vehicles currently being produced. While that's no big deal to you, it is to some of us -- and there's much more than a Tesla event going on here. We've been around long enough to watch all this stuff evolve from inception to fruition so saying it's not here is just ridiculous to some of us.

 

[RubberToe]

Can we wait until it's here before we start claiming it's here? It's like a Tesla release event in here!

Wait a minute, somebody got it? I didn't get it! why did he/she get it and I didn't? wheres that website tracking firmware releases? Who's that guy still running 5.17, how did that happen? Why is my hardware obsolete now? Someone over there said it can be retrofitted on my old 2013 Model S! Does it also support battery swap on the fly? If the factory started production on 10-11, painted on 10-13, Elon inspection on 10-17, and then loaded on the ship to Botswana on 10-20, do you think it has the EAP hardware? My delivery specialist in Nigeria (same guy with those cute puppies!) said I could upgrade in route for only $3,000 USD if I give him my bank account number! Anyone want to do a group buy on this? Gotta move quick though, only so many retrofits can be done on that ship, thats what my DS says. Don't believe those DS's, they don't know anything compared to the community knowledge base here! Wait a minute, someone else got the first one over there! But the USB port doesn't support my peta-byte multi-partitioned SATA global music archive with 138,099,988 songs, all with embedded TRACKARTISTINSTRUMENT cover art! What were they thinking? This new EAP release is DOA, I'm sitting on my 6.23.878 GOLDEN release until 9.26.998 fixes all this nonsense! Tesla will not sell a single car until this all gets corrected!!! Damn, I really wish I had waited 1 more freaking day to order the car, who knew Tesla would just up and upgrade stuff!!! Barista! One more octuple espresso please! Gotta find out who got it. PLEASE post fully coreographed and multi-lingual documented first drive video ASAP! And, most importantly, will the cup holders hold a 7-11 64oz. small bottomed slurpee cup, not the new ones, the older ones that my store never upgraded from...

Just trying to bring some sanity back to the thread

RT