Seeing the world in autopilot, part deux

[Bladerskb]

No one uses mobileye products in retail cars. Stop pushing the lies. It's not Beach Week. Mobileye only has demos. Audi L3 was as much of a joke as I thought it was. No one else uses anything because it's not ready or other companies are too chicken *sugar* to have confidence in the tech.

huh what does AP1 and supercruise (just to name two) run on?


this is a joke right?

 

[croman]

huh what does AP1 and supercruise (just to name two) run on?


this is a joke right?

You said it yourself, 4 year old tech. Not relevant.

 

[Bladerskb]

Again you a comparing a 4 years old product with a 2018 product that only finally today (with 9.0) is able to match/eclipse the 4 years old product in driving output performance and features. Although they are lacking behind in other detection networks that eyeq3 has.

Actually i take it back. the amount of false positive (for pedestrian for example) is crazy..
Would like to see if that improved in 9.0 before making any final judgments.

 

[verygreen]

You won't know that without orientation, though, if they are stopped.

By definition if they are stopped, they don't have a direction. And from a stopped position car can move forward, backwards and somewhat sideways.

 

[verygreen]

Every single SDC company are research labs (including waymo) other than Mobileye. Why? Mobileye is actually SELLING their products. No othe company is selling complete SDC hardware and software today other than Mobileye. Let that sink in.

This isn't some demo. This is real!

I am pretty sure Tesla is selling (claiming to, anyway) SDC hardware and software since... I dunno, Oct 2016?

On the other hand I already asked you, what dealership do I go to TODAY to try the superior solution from Mobileye?

Mobileye product ARE actually REAL! Prove in point is AP1 and Super cruise using 4 year old eyeq3.
Think about it, Mobileye was 4 YEARS AGO where Tesla is today.
You think the last 4 years mobileye has just been sitting on its ass doing nothing?

So what dealership was that, again?

they don't show demo ware, they don't show research projects, they don't show anything that doesn't have a production deal. Eyeq4 is IN production. BMW cars are getting it this year. Nissan is saying they will have LVL3 using eyeq4 early 2019.

So... Nowhere I can go TODAY? I see. Please remind me once I can go see something with my own eyes.

This is a huge difference between mobileye vs other companies, they actually have to sell stuff to make money.

Make money? They already made their money and you seem to have forgotten that Intel bought them quite awhile ago and so they don't "need to make money" in the normal sense anymore.

I do wonder how long until Intel spins them off/sells them to somebody else, though. Then your assertions might become true again (which is not a given since they might sell them to some particular vendor too).

 

[S4WRXTTCS]

I just looked, and GDOT page says that HOV lanes use double white lines to indicate no crossing, not just a single line. This follows the federal guidance around lines. Double white is prohibited cross, single while is restricted crossing. For Tesla to obey the laws, they need to track if it's a single or double line.

I agree that auto lane change can't cross solid white all by itself as it doesn't understand the restriction. I just want it to cross when I ask it too, instead of pretending that lane doesn't exist at all.

In the V9 firmware they have a setting to instruct it how to handle HOV lanes with the auto-lane changer thing.

So I'm curious if they've added additional smarts like knowing what lane is an HOV lane, and where it's okay to cross.

I have a strong feeling like a lot of things are coming together with V9.

Better Maps
Blindspot monitoring that uses the side cameras
Reorganized IC to show cars in front, and on the sides.

I can't wait to try it on 405 with the double HOV lanes, and restricted entry.

 

[thegruf]

@Bladerskb unsurprising to see you chip in with you unwaveringly negative view on anything Tesla.

You predictably miss one key point in your analysis compared to mobileye.

What @verygreen has (most excellently) posted is their unofficial interpretation of the rawest information available from the Tesla AP, you have absolutely no idea what further processing Tesla apply to this and nor do I, yet you compare this raw unoffical data to an official presentation from Mobileye, by Mobileye people with absolutely no knowledge about how much that video has been postprocessed.

I don't doubt for a moment that what Mobileye are doing is right up there with the best (except for Waymo who seem to be in a class of their own), what I also see is Tesla delivering vehicles today with more advanced AP features than anything else currently widely available, and Tesla with their own roadmap, rate of development and internally developed hardware on the way, every reason to believe Tesla will be the standard by which others are measured for some time to come.

Everyone has something new and better coming soon ... including Tesla.

 

[chispas]

@Bladerskb even if you are right and Tesla's tech is at parity with something that is 4 years old they have achieved parity with a company that has been doing this since 1999 in only 2 or 3 years. If you extrapolate that, how long until Tesla catches-up with what they have built since then? 1 year? 2 years?

 

[Bladerskb]

I am pretty sure Tesla is selling (claiming to, anyway) SDC hardware and software since... I dunno, Oct 2016?

On the other hand I already asked you, what dealership do I go to TODAY to try the superior solution from Mobileye?


So what dealership was that, again?


So... Nowhere I can go TODAY? I see. Please remind me once I can go see something with my own eyes.


Make money? They already made their money and you seem to have forgotten that Intel bought them quite awhile ago and so they don't "need to make money" in the normal sense anymore.

I do wonder how long until Intel spins them off/sells them to somebody else, though. Then your assertions might become true again (which is not a given since they might sell them to some particular vendor too).

The GM dealership. The GM supercruise makes Tesla autopilot look like a toy. This is the exact quote from geohut whom alot of ppl here look up to. (He owns a supercruise) 6+ other journalist who reviewed both at the same time say sc destroys Tesla ap2.

 

[Snuffysasa]

In the V9 firmware they have a setting to instruct it how to handle HOV lanes with the auto-lane changer thing.

So I'm curious if they've added additional smarts like knowing what lane is an HOV lane, and where it's okay to cross.

I have a strong feeling like a lot of things are coming together with V9.

Better Maps
Blindspot monitoring that uses the side cameras
Reorganized IC to show cars in front, and on the sides.

I can't wait to try it on 405 with the double HOV lanes, and restricted entry.

Hey I like the optimism. about v9.

I do not know what Tesla does. but I do know there are some map companies that makes maps for autonomous driving, and they do have additional smarts like knowing the exact start and end locations of HOV lanes and where it ok to cross.

 

[croman]

The GM dealership. The GM supercruise makes Tesla autopilot look like a toy. This is the exact quote from geohut whom alot of ppl here look up to. (He owns a supercruise) 6+ other journalist who reviewed both at the same time say sc destroys Tesla ap2.

LOL. That is quite possibly the silliest thing you've ever said. 0 credibility. The General Assembly sends its fullest guffaws.

GM Supercruise is so limited. Clearly its having an impact in sales, right? Those GM luxury cars must be outselling Tesla because it has the BEST ADAS EVA!

The reason Geohot owns Supercruise is because Tesla blocked him from buying an AP2 car. He's not exactly unbiased nor that knowledgeable about Tesla's developmental state.

The answer is that MobilEye might have great demos, but no major auto manufacturer is using EyeQ4 and the plans for its use are just that -- plans and nothing concrete. Tesla has 1-2 years of further development before anyone even thinks of selling EyeQ4 based ADAS (or SDC). Also these auto companies need to actually write and test software. Few have even 1/10th the talent Tesla currently employs for this task. Finally, Tesla is gathering data that these companies will need to spend years accumulating.

 

[Snuffysasa]

The answer is that MobilEye might have great demos, but no major auto manufacturer is using EyeQ4 and the plans for its use are just that -- plans and nothing concrete.

This is simply not true.

GM Supercruise is so limited. Clearly its having an impact in sales, right? Those GM luxury cars must be outselling Tesla because it has the BEST ADAS EVA!

This is a silly argument... I could make the same argument, that there are no Waymo cars I can buy today. Therefore, Waymo must not have as good of a system as Tesla...

And as I'm sure you know, people are buying Teslas over Cadilacs for dozens if not hundreds of reasons not including the ADAS system

 

[verygreen]

The GM dealership. The GM supercruise makes Tesla autopilot look like a toy. This is the exact quote from geohut whom alot of ppl here look up to. (He owns a supercruise) 6+ other journalist who reviewed both at the same time say sc destroys Tesla ap2.

ok. what particular car do I ask them for? I'll try it on Saturday. Keep in mind I live in the boonies, and I think there was some NY guy even that complained that due to no maps in his area the system barely activated.

Keep in mind I'll be comparing against v9 too

 

[strangecosmos]

We have to compare apples to apples and oranges to oranges. Compare demos to demos and compare field tests to field tests. Don’t compare demos to field tests.

 

[Cirrus MS100D]

Keep in mind I'll be comparing against v9 too

You’re a goddamned tease! I’m assuming you’re running v9 already?!?!

 

[S4WRXTTCS]

The GM dealership. The GM supercruise makes Tesla autopilot look like a toy. This is the exact quote from geohut whom alot of ppl here look up to. (He owns a supercruise) 6+ other journalist who reviewed both at the same time say sc destroys Tesla ap2.

I'd say I'm someone who would vastly prefer Supercruise simply because of the hands free, and the way GM constantly updates the maps by using lidar equipped mapping vehicles.

For L2 driving on white listed roads there isn't anything better at this time. But, you and I know that's not really because of MobileEye that makes it special.

There is nothing stopping Tesla from doing the same, and in fact Engineers at Tesla wanted to use something similar for driver monitoring that GM uses in Supercruise.

Elon overruled that, and that's why were stuck with the silly steering wheel torque sensor.

But, your issue is you can't be excited by the different offerings various companies have. You can't though because you just have to continue being negative instead of understanding that there are trade offs.

Right now I'm like a kid before Christmas anxiously awaiting the release of V9. Do I really think it's going to work in average Seattle conditions? Probably not. Do I have lots of disagreements about the implementation? Yes.

But, I can find excitement in it despite any misgivings I might have.

At the end of day I'm excited by autonomous technologies. You seem to be excited by nit picking, and bickering.

 

[verygreen]

You’re a goddamned tease! I’m assuming you’re running v9 already?!?!

Well, the car is at home and I am stuck in Paris for a couple more days. But you better believe I'd be running v9 once I get back. The internals are interesting and it's quite a rewrite of the AP subsystem. Does use all cameras it appears

 

[AnxietyRanger]

You won't know that without orientation, though, if they are stopped.

By definition if they are stopped, they don't have a direction. And from a stopped position car can move forward, backwards and somewhat sideways.

In theory.

In practice, the orientation of stopped cars in traffic is used very effectively by everyone of us on a daily basis, we don't even think about it probably. If a car is stopped, facing traffic in an intersection, it is probably waiting to merge or cross - and we keep an eye on that. If it is facing the other way around, it is probably waiting to park or drive away etc., we hardly expect them to back into the traffic... could happen, but the probabilities of it can help autonomous vehicles make similar judgement calls.

Also one can estimate potential turning arcs based on the angle of a stopped car, no matter which direction it takes. Knowing the orientation of a car makes a massive difference in this, compared to simply knowing a car is there.

Look, I'm the first to admit Tesla's more aggressive approach to self-driving development can certainly yield consumer-facing results faster than the competition. And in some areas, auto-steering especially, it has. But IMO that is separate from analyzing their vision prowess. Who has the best vision and who does what with their vision are two separate questions. What this thread IMO is mostly about is the vision part, since that is what your super-great findings provide us "vision" into.

Using that vision to do stuff is separate: AP1 is actually a good example of this. A lot of companies use MobilEye's vision chips, but only Tesla used EyeQ3 so aggressively with AP1. Had Tesla put a fisheye camera as a second camera (EyeQ3 supports several cameras), they could have done Autosteer+ with it, done steep hills with it (without the narrow camera losing sight of the road), and still have traffic-sign recognition and all that, years ago. Then again, other companies with very small and conservative ADAS teams used the same chip(s) much less effectively even in the AP1 era and are only now beginning to reach those AP1 levels in consumer products...

IMO there is little reason to doubt MobilEye's vision surpasses that of Tesla's - and I must admit I'm surprised you seem to think so and didn't see how obviously the narrow AP1 camera issue being the reason it can't see hills or tight curves. Personally, I find it likely EyeQ3 is already superior to current AP2+ in vision with similar cameras and even more likely that EyeQ4 is significantly ahead... @Bladerskb does have a point: MobilEye sells these as chips to OEMs to use, they must be finished products. MobilEye also has a very long and public academic history of research and showing their results in those settings. IMO there is little reason to doubt MobilEye's vision capability.

The interesting question here IMO isn't who is leading vision (that would probably be a debate between Waymo and MobilEye), it is how quickly can MobilEye's (or Waymo's, if they so choose) partners use that vision ability and create effective results the consumer can buy and use. That is IMO Tesla's real edge (and a bit similarly Comma.AI's edge) - they obviously are far more willing to run with more uncertain results and ship that to consumers. As a consumer, getting faster results with Tesla seems definitely possible. Here is where, say, I and @Bladerskb differ: I am definitely excited about the potential of this and do acknowledge a meaningful difference in approaches. I look forward to AP2 V9 (even though I'm apprehensive about the UI changes based on past experiences).

Other car-makers are much more conservative in their approach. Then again, others are more willing to use additional sensor technologies to help make their progress in other ways, including lidar and radar, which could also make some difference. I also find it possbile Tesla will be left behind in actual Level 3-4 progress (driver not responsible), while offering similar or better progress on Level 2, where driver still remains responsible for example... which would give the competition another kind of edge, of course. We shall see how it all plays out... Interesting times.

Thank you @verygreen and @DamianXVI for the good work in bringing us a glimpse of what Tesla's vision status has been recently!

Keep up the good work!

 

[Bladerskb]

LOL. That is quite possibly the silliest thing you've ever said. 0 credibility. The General Assembly sends its fullest guffaws.

GM Supercruise is so limited. Clearly its having an impact in sales, right? Those GM luxury cars must be outselling Tesla because it has the BEST ADAS EVA!

Again 6+ independent journalists who pitted supercruise against autopilot said SC destroyed it.

The reason Geohot owns Supercruise is because Tesla blocked him from buying an AP2 car. He's not exactly unbiased nor that knowledgeable about Tesla's developmental state.

The diff between u and me is that my statements are based on actual facts. George is like the biggest elon hugger. He literally worships elon in every periscope. I know because I have watched almost all of his relevant videos. Unlike alot of people here I dont live in a Tesla bubble.

He thinks that all other companies is a scam and that Tesla will win self driving and will be the apples of cars and thinks elon is the second coming of Jez! So for you to accuse him of being biased against Tesla is laughable!

He uses supercruise every day and he has extensively tested ap1 and ap2 and he says "super cruise is the best autonomy you can buy today"

He says his own openpilot wobbles and autopilot wobbles but supercruise is "dead center like a car on rails" he also said "I have never seen supercruise make a mistake"

He says the way SC steers is something you have to experience to believe and that its better than everything else out there.

Also he said they can hack and make SC to show up in any road. They already had the supercruise ACC working. So the geofencing of SC is by design not some huge limitation.

Supercruise on the highway/freeway is like gawd mode. Also some people dont realize but GM has already done several software update for supercruise. For example heavily improving the ACC and then reducing the activation constraints so now SC shows up readily compared to before. Basically all the complaints about supercruise is gone.

GM Makes Super Cruise Driver-Assist System Easier to Use

The answer is that MobilEye might have great demos, but no major auto manufacturer is using EyeQ4 and the plans for its use are just that -- plans and nothing concrete. Tesla has 1-2 years of further development before anyone even thinks of selling EyeQ4 based ADAS (or SDC). Also these auto companies need to actually write and test software. Few have even 1/10th the talent Tesla currently employs for this task.

You are simply out of your element.

Finally, Tesla is gathering data that these companies will need to spend years accumulating.

They dont need to, they have mobileye rem mapping already gathering data.

 

[croman]

@Bladerskb - I said George had biases not that he's biased against Tesla.

The rest of the comments are just a rehash of prior disagreements. Your haughty dismissals do nothing to move the needle. Supercruise is too geofenced to actually supplant AP for me. To each their own.

Tesla has the balls to actually push the envelope. I don't like their business practices but at least they are trying. I don't see major auto companies doing the same in a similar timeframe despite your incessant ME cheerleading.

Which current car can automatically change lanes without driver input? Supercruise can't even do what Teslas have been doing for 4 years now with turn signal lane changing. Your narrative you push is getting stale.