Elon tweets "pure vision" will solve phantom braking

[gearchruncher]

Yup, that's why Tesla must implement some cloud maps feature, but it's contingent on vision achieving very high accuracy.

I thought Tesla was going with first principles, vision only, lIke a human drives, and that HD maps are a crutch for less advanced systems lIke Cruise and Waymo with their geofences. Elon literally called them a bad idea, so there is no way they are using them.
If the first car can successfully navigate by itself, so can the second. No need for the first to send a map to the second. Construction zones can amd do get changed hourly, the bandwidth needed would be immense.

 

[mikes_fsd]

And now, kids, we're going to go into the HD Maps rabbit hole!

3-2-1 go

 

[gearchruncher]

And now, kids, we're going to go into the HD Maps rabbit hole!

Yeah, not like someone that is very positive on Tesla's self driving abilities said:

Yup, that's why Tesla must implement some cloud maps feature, but it's contingent on vision achieving very high accuracy. As any Tesla cars pass by road conditions, they'll be updating the maps, and we all benefit.

In direct contradiction to Elon saying:

High precision maps and lanes are a really bad idea ... any change and it can't adapt.

You don't get to throw that out there and then act like other people are bringing up HD maps.

So which one is it? Does Elon know what he is doing, or do you guys know better and know maps are key to good phantom braking performance?
I assumed this was solved- The Technoking told us it wasn't a good idea, it wasn't first principles (just like LIDAR isn't), and so it isn't needed. Yet now we're saying it's a "must?".

 

[Tam]

And now, kids, we're going to go into the HD Maps rabbit hole!

3-2-1 go

HD Mapping: Friend or Foe of Robocars?​

The above article says that Elon Musk is partially right "During the company’s Autonomy Day in April, Musk made it abundantly clear that too much dependency on HD Maps can turn an autonomous vehicle (AV) into a “system that becomes extremely brittle,” making it more difficult to adapt."

But they add that when using HD maps correctly, it's life-saving if it was used in the Mountain View 2017 Model X Autopilot accident in March 2018: The camera mistook the different colors in the darker asphalt surface and the lighter concrete surface as a lane line which lead the car right straight into the concrete barrier:

 

[emmz0r]

https://twitter.com/x/status/1393389199412416517

Here's another one. If this mode remains and highway driving does not use the stitched model, this does NOT instill confidence lol

 

[raptor5244]

At this point I would be happy with Rear Cross Traffic Alert.

 

[Bladerskb]

If it's a map issue, then probably not. I think once Tesla Vision(TM) achieves high accuracy, Tesla will deploy their own cloud-based mapping system, allowing cars to automatically send mapping updates to the cloud (like Mobileye REMS but better).

If someone is doing something that Tesla isn't doing then its stupid, useless and a crutch.
But when Tesla does it, its amazing, a quantum leap.

Even worse, when someone else is working on something it means nothing and shouldn't be talked about because its not here, its in the future so its pointless.
But when it comes to Tesla. Even if you don't know if they are working on something and exactly the details of it. Hype it up and say its better than what others are already doing right now which you also call stupid, useless and a crutch.

Rinse and Repeat.

 

[powertoold]

If someone is doing something that Tesla isn't doing then its stupid, useless and a crutch.
But when Tesla does it, its amazing, a quantum leap.

Cloud maps is an obvious development following accurate vision. Unfortunately, Mobileye's vision isn't as good as Tesla's. I think REMS is based on the 2018 tech / chips / NN. It's why Bmw's traffic light feature has yet to be deployed. And also why Mobileye's production autosteer implementations are much worse than Tesla's.

 

[rxlawdude]

If someone is doing something that Tesla isn't doing then its stupid, useless and a crutch.
But when Tesla does it, its amazing, a quantum leap.

Even worse, when someone else is working on something it means nothing and shouldn't be talked about because its not here, its in the future so its pointless.
But when it comes to Tesla. Even if you don't know if they are working on something and exactly the details of it. Hype it up and say its better than what others are already doing right now which you also call stupid, useless and a crutch.

Rinse and Repeat.

Complain and whine. Rinse and repeat. Again and again. Incessantly.

 

[gearchruncher]

Cloud maps is an obvious development following accurate vision.

Just one that the Technoking has called a "really bad idea." You'd think if it was obvious, he'd know.

Replace "LIDAR" with "Cloud Maps" in this quote from Kaparthy and tell me why one is needed but the other is a bad idea:

"Lidar is really a shortcut," added Tesla AI guru Andrej Karpathy. "It sidesteps the fundamental problems of visual recognition that is necessary for autonomy. It gives a false sense of progress, and is ultimately a crutch."

None of this is complaining. It's just pointing out that Tesla is all over the place and inconsistent with their tech and historical statements, which means we should be highly skeptical about their future performance until we see it. Tesla thought they needed Radar ("A forward-facing radar with enhanced processing provides additional data about the world on a redundant wavelength that is able to see through heavy rain, fog, dust and even the car ahead.") until they didn't. They said HW2 was all they needed until they needed more.

So who knows if moving to pure vision will make things better or worse on average, since we have no reason to put a lot of credence into what they say. It's confusing that some of you are so sure that we will be getting improvements just because Elon tweeted we would.

I do know one thing. Using maps is not "pure vision" so if they are using that, it's yet another place where Elon isn't being honest in his communications.

 

[powertoold]

I don’t respond to you about maps cuz you seem to misunderstand the difference between hd maps and a normal map.

 

[gearchruncher]

If it's a map issue, then probably not. I think once Tesla Vision(TM) achieves high accuracy, Tesla will deploy their own cloud-based mapping system, allowing cars to automatically send mapping updates to the cloud (like Mobileye REMS but better).

Yup, that's why Tesla must implement some cloud maps feature, but it's contingent on vision achieving very high accuracy. As any Tesla cars pass by road conditions, they'll be updating the maps, and we all benefit.

I don’t respond to you about maps cuz you seem to misunderstand the difference between hd maps and a normal map.

No, you don't respond because you clearly mean maps of high enough quality that they actively are part of the driving task (not just navigation) and this is not "pure vision". You don't need to update normal maps for construction that shifts a lane over, closes one, or pops up a few cones because humans can handle that just fine.

If you mean just navigation, why didn't you just respond "Tesla doesn't use maps for braking, that would be HD maps" instead of going down this trail of how they must implement maps like REMS?

 

[Bladerskb]

Cloud maps is an obvious development following accurate vision. Unfortunately, Mobileye's vision isn't as good as Tesla's. I think REMS is based on the 2018 tech / chips / NN. It's why Bmw's traffic light feature has yet to be deployed. And also why Mobileye's production autosteer implementations are much worse than Tesla's.

HD maps are a joke imo. Elon is making more and more sense as time goes by. HD maps are great for demos and raising capital. They seem suited for creating standardized routes for freight or mass transit. They'll never get to level 5 with HD maps. Using HD maps essentially locks you into level 4 by definition. HD maps aren't even good training wheels for FSD. Their purpose is to trick people (and investors) into believing autonomous cars are coming soon. These silly companies have been tinkering with HD maps since 2017 or before and still haven't come with a single profitable route or region or mass transit application. It's a farce.

HD maps are safer for sure, if properly created and maintained.

Karpathy doesn't think HD maps are scalable, and even if they are scaled, they need to be maintained, which may be either cost prohibitive or not reliable enough given human limitations. The funny thing is that many companies are using vision to create rough HD maps, which begs the question: once your vision is good enough, why even use HD maps?

HD maps will likely turn out to be the fuel cells of FSD. That's essentially what Karpathy is saying. No one knows the future for certain, but companies without fleet data are forced to use HD maps. You guys need to have that sink in. If they had the fleet data AND still decided to use HD maps, then it'd be a different story.

Without HD maps, they can't make fsd safe because they don't have free trainers like Tesla has.

One of the biggest challenges of HD maps is deciding when the vision will be prioritized over the HD maps. For example, if a portion of the highway lanes is being repainted and the vision says one thing but the HD maps says another. When do you use the vision? It kind of begs the question, if vision is so important and needs to be a certain level of accuracy and reliability, why even use other sensors? Why not just use redundant vision?

Cloud maps is an obvious development following accurate vision. Unfortunately, Mobileye's vision isn't as good as Tesla's. I think REMS is based on the 2018 tech / chips / NN. It's why Bmw's traffic light feature has yet to be deployed. And also why Mobileye's production autosteer implementations are much worse than Tesla's.

Additionally, Is that why Mobileye will deploy driverless autonomous fleet before Tesla? The EyeQ4 was years ahead of Tesla Vision in NN development and deployment. Not only does Mobileye use the EyeQ4, but also the upcoming EyeQ5 will be apart of the REM fleet. Also the BMW Traffic light feature is already deployed. Lastly there is no single Mobileye production autosteer implementation. Supervision is the very first and will be on the 2021 Zeekr 001.

 

[rxlawdude]

The tag team of Whiners is here and taking aim.

 

[raptor5244]

It's a good thing this car is fun to drive on our own.

 

[powertoold]

Also the BMW Traffic light feature is already deployed. Lastly there is no single Mobileye production autosteer implementation.

No, it's not deployed. Stop gaslighting, and all the Mobileye production autosteer implementations are much worse than Tesla's.

Mobileye will fail. Intel made a poor decision. Amnon is a slick talker. Zeeker 001 will have low sales, less than 20k a year I predict.

 

[Bladerskb]

No, you don't respond because you clearly mean maps of high enough quality that they actively are part of the driving task (not just navigation) and this is not "pure vision". You don't need to update normal maps for construction that shifts a lane over, closes one, or pops up a few cones because humans can handle that just fine.

If you mean just navigation, why didn't you just respond "Tesla doesn't use maps for braking, that would be HD maps" instead of going down this trail of how they must implement maps like REMS?

They are a walking contradiction.

 

[Bladerskb]

No, it's not deployed.

It is deployed in Germany and certain locale, it also doesn't use Mobileye's REM.

Stop gaslighting, and all the Mobileye production autosteer implementations are much worse than Tesla's.

Mobileye will fail. Intel made a poor decision. Amnon is a slick talker. Zeeker 001 will have low sales, less than 30k a year I predict.

That's a contradiction. Tesla's AP1 also uses mobileye.

Supervision is the first time Mobileye has done the driving policy and control for a system. They usually only provide the perception chip and software alone. Which Tier 1s like Aptiv, ZF, etc & automakers like GM, Ford, Nissan, Tesla (AP1), NIO, etc, develop driving policy and control algorithms from.

"This win marks the first time Mobileye will be responsible for the full solution stack, including hardware and software, driving policy and control." - Mobileye

Mobileye will fail. Intel made a poor decision. Amnon is a slick talker. Zeeker 001 will have low sales, less than 30k a year I predict.

While you wish others fail. I for one can't wait for Mobileye's supervision and Huawei's Advanced Autopilot. I wish i could buy them all so i can drive them everyday and compare.

 

[powertoold]

While you wish others fail. But I for one can't wait for Mobileye's supervision and Huawei's Advanced Autopilot. I wish i could buy them all so i can drive them everyday and compare.

China will definitely leapfrog Mobileye quickly (if they haven't already). I don't doubt that. I'll wait and see regarding SuperVision, but at this point, v9 will come out before the first SuperVision car is produced.

 

[Bladerskb]

China will definitely leapfrog Mobileye quickly (if they haven't already). I don't doubt that. I'll wait and see regarding SuperVision, but at this point, v9 will come out before the first SuperVision car is produced.

Huawei already said they are ahead of Tesla and it shows with what they are capable of doing in an environment that is orders of magnitude harder to drive in than in the US.

"He further commented on one of the giant in the electric vehicle industry, Tesla, saying that the Huawei team is doing its best, and even if the team doesn’t brag about it but it’s true that Huawei can achieve 1000 kilometers of uninterrupted autonomous driving in the urban area, which is way better than the U.S. electric car maker, Tesla."

Better than Tesla, Huawei cars provide 1000 Kilometers of non-stop autonomous driving

On April 12, at Global Analyst Summit, Huawei Rotating Chairman, Eric Xu, said that Smart Car Business is an industry that Huawei focuses on and it’s a business with complete features. Huawei has invested over 1 billion US dollars in smart car components this year, along with its investment in...

www.huaweicentral.com

What Huawei and Mobileye primarily prove is that there is absolutely nothing special about what Tesla is doing. That they have a general system that is more reliable in an even more complex environment and road system which we will see in a few months.

This nullifies many myths that were already false to begin with.
Example:

• You need billions of miles from millions of cars to develop the AP, NOA and FSD Beta system that Tesla has
• Therefore tesla is years ahead because they have more data.

but at this point, v9 will come out before the first SuperVision car is produced.

The faithful as always will profess and cling to the next shiny object. This time its version 9. Next time it will be 9.1, 9.2, 9.4, 9.5, then 10.
Then it will be wait till Dojo, then wait till HW4, it never ends.

You famously said that in 6-9 months they would be at around 150k miles per safety disengagement.
Its been almost 7 months and they are still at 1-5 miles (on average) per safety disengagement.

Fun fact: at Autonomy Day, they were at safety disengagement per drive on average according to analysis of multiple people having been in rides that required a safety disengagement. Note that these were easy routes and wide roads with almost no traffic. 2 years later, they are still at safety disengagement per drive on average.

But ofcourse V9 is the holy grail that will beset upon us Level 5.