Autonomous Car Progress

[nativewolf]

Mobileye is not going the Tesla-like approach. One, Mobileye has not changed their approach. Two, Mobileye has always believed in this approach focused on vision-only + maps + RSS for hands-off and imaging radar and lidar for eye's off. That is not the Tesla approach.

Sure, both Mobileye and Tesla are focused on consumer cars and going from L2 to L4 and both have a vision-only L2 product. But that is where any similarities end. If you actually look at the two approaches, they are very different.

Mobileye uses 11 HD cameras, maps and RSS and hands-free. Mobileye also adds lidar and imaging radar for eye's off. And Mobileye has a robotaxi.
Tesla uses 8 cameras, no maps, no RSS, no radar, no lidar. And it is hands-on. And Tesla has no eye's off. Tesla has no robotaxi.

It is more than just maps that are different. The cameras are different, the use of RSS, the use of radar and lidar are also different between Mobileye and Tesla.

So how is Mobileye going the Tesla-like approach when there are so many differences?

To say that Mobileye is going the Tesla-like approach is preposterous.

Methinks he is just trolling you guys

 

[diplomat33]

Mobileye demos their lidar that is small and cheap for consumer cars. The lidar can measure velocity as well as distance and shape of objects.

https://twitter.com/x/status/1611242911663886337

 

[diplomat33]

I could not imagine using Supervision without their REM maps. The part of the Mobileye CES video (below) where they talked about how quickly they were able to populate their REM maps with real-world data was unbelievable in its breadth. And the "tiny" amount of data/bandwidth that was required. No video just the 'facts'.

Mobileye’s Approach – AV Maps

• Scalable-by-design: Millions of Mobileye-equipped ADAS vehicles sending data to the cloud in small data packets (10kb/km)
• Fully automated map generation at the push of a button
• Maps can be updated in near-real time because of sophisticated change-detection algorithms on millions of mapping agents
• Superior local accuracy: where it matters
• Using the “wisdom of the crowd” to create rich semantic layer of driving culture and traffic rules

Mobileye REM™ - Road Experience Management​

2058 seconds in or 34:18 minutes

Yeah, Mobileye's REM maps are awesome.

 

[scottf200]

I was curious about which 17 brands Mobileye is getting data from as well as partnering with.

The chips are of wikipedia has some of it in the last row: Mobileye - Wikipedia

 

[Matias]

Mobileye CES talk:

IMO, it was super interesting.

Things I loved:

Mobileye defined consumer based "levels of autonomy" and ODD.
They showed us the exact sensors, ODD, compute etc of all their AV products.
They have a path to scale from consumer to robotaxi. They have a path to scale ODD.
They explained how they validate safety and what their safety goal is.
They showed some good examples of SuperVision handling difficult cases.
They have clear business model and timeline for both consumer cars and robotaxis.
Redundancy in sensors

1:12 “full self driving car hardware cost below 5000$”. And he is talking about real full self driving. If OEM would sell it with 10 k, I would buy it.

 

[Matias]

That was an interesting watch.
Their confidence makes me hesitate because it's impossible to be that confident this early on. At this point, both tesla and Mobile eye, both don't know what they don't know. They are technically just predicting thry will solve all these issues eventually.

They have been working on this field a lot longer than Tesla. You might now, that Tesla’s AP1 was based on their chip.

 

[Matias]

Thanks for that link. So it looks like we might get SuperVision in the US later this year if Zeekr does bring the Zeekr001 to the US as planned.

As my Model S85D next summer reaches 8 years and warranty expires I’ve been planning to buy Model Y, but now I noticed that Zeekr plans to launch in Europe this year

 

[diplomat33]

1:12 “full self driving car hardware cost below 5000$”. And he is talking about real full self driving. If OEM would sell it with 10 k, I would buy it.

Definitely, I would too.

As my Model S85D next summer reaches 8 years and warranty expires I’ve been planning to buy Model Y, but now I noticed that Zeekr plans to launch in Europe this year

If I were you, I'd wait a bit longer and check out Zeekr first and see if it is a good match for you.

 

[powertoold]

Mobileye is not going the Tesla-like approach. One, Mobileye has not changed their approach.

Maybe I'm wrong, but this is the first time I've seen Mobileye talk about their "shadow mode" and generative "simulation" approaches.

"Shadow mode" is a Tesla term from around 2018, and it connotes a "data engine"-type iterative approach with a large fleet, which is the heart of Tesla's approach.

Generative simulation was shown at Tesla's AI Day 2, but Tesla seems much further ahead both in simulation and "shadow mode."

That's why I said that Mobileye seems to be adopting a Tesla-like approach.

 

[joenathan]

Maybe I'm wrong, but this is the first time I've seen Mobileye talk about their "shadow mode" and generative "simulation" approaches.

"Shadow mode" is a Tesla term from around 2018, and it connotes a "data engine"-type iterative approach with a large fleet, which is the heart of Tesla's approach.

Generative simulation was shown at Tesla's AI Day 2, but Tesla seems much further ahead both in simulation and "shadow mode."

That's why I said that Mobileye seems to be adopting a Tesla-like approach.

Lol, a "Tesla term" Deploying Machine Learning Models in Shadow Mode

 

[diplomat33]

Maybe I'm wrong, but this is the first time I've seen Mobileye talk about their "shadow mode" and generative "simulation" approaches.

"Shadow mode" is a Tesla term from around 2018, and it connotes a "data engine"-type iterative approach with a large fleet, which is the heart of Tesla's approach.

Generative simulation was shown at Tesla's AI Day 2, but Tesla seems much further ahead both in simulation and "shadow mode."

That's why I said that Mobileye seems to be adopting a Tesla-like approach.

Mobileye and Tesla do not use shadow mode the same way. Tesla uses shadow mode for development. Mobileye uses shadow mode only for validation and not development. That is a key difference.

And Tesla did not invent shadow mode.

Also, Mobileye and Tesla do not use generative simulation the same way either. Mobileye generates simulation from their REM maps which they have crowdsourced from their fleet around the world. Tesla does not.

There are many other key differences between the two approaches that I pointed out in my previous post.

So no, just because Mobileye used two terms that Tesla also used in a presentation does not mean that Mobileye is adopting a Tesla-like approach. How they do shadow mode and simulation is different. And there are many other differences. Mobileye is not following a Tesla approach.

 

[powertoold]

Mobileye generates simulation from their REM maps which they have crowdsourced from their fleet around the world. Tesla does not.

I'm not saying Mobileye IS USING Tesla's approach. I'm saying Mobileye seems to be adopting a Tesla-like approach.

As for REM vs Tesla crowdsourcing, see here:

Tesla AI Day 2022

Join us to build the future of AI → https://www.tesla.com/ai--------0:00:00 - Pre-show0:13:56 - Tesla Bot Demo0:29:15 - Tesla Bot Hardware | Hardware Archite...

www.youtube.com

 

[diplomat33]

I'm not saying Mobileye IS USING Tesla's approach. I'm saying Mobileye seems to be adopting a Tesla-like approach.

As for REM vs Tesla, see here:

Tesla AI Day 2022

Join us to build the future of AI → https://www.tesla.com/ai--------0:00:00 - Pre-show0:13:56 - Tesla Bot Demo0:29:15 - Tesla Bot Hardware | Hardware Archite...

www.youtube.com

It is not a Tesla-like approach. There are too many differences.

Just because mobileye used two words that Tesla also used on AI Day, does not make it Tesla-like.

 

[powertoold]

Just because mobileye used two words that Tesla also used on AI Day, does not make it Tesla-like.

Well we can agree to disagree, but the core of Tesla's approach is a large fleet with an iterative data engine and shadow mode. Mobileye is trying to gather a large fleet (they say 60k+ Zeekrs now), and now they have the fleet to do shadow mode, OTAs, and simulation.

Whether you say the shadow mode is for development or validation is potato potato (Tesla uses it for validation and development).

 

[2101Guy]

Methinks he is just trolling you guys

I just read the last few pages. Diplomat33 DESTROYED dude in the waymo debate. Diplomat was Gervonta and KS was Hector Garcia.

 

[Bladerskb]

Maybe I'm wrong, but this is the first time I've seen Mobileye talk about their "shadow mode" and generative "simulation" approaches.

"Shadow mode" is a Tesla term from around 2018, and it connotes a "data engine"-type iterative approach with a large fleet, which is the heart of Tesla's approach.

Generative simulation was shown at Tesla's AI Day 2, but Tesla seems much further ahead both in simulation and "shadow mode."

That's why I said that Mobileye seems to be adopting a Tesla-like approach.

Lol this is wrong.

Tesla is literally behind on simulation compared to almost everyone else who went headlong into it.
Tesla was the biggest advocate AGAINST simulation and did a u-turn on AI Day and started copying everyone else.
Dude you need to get out of your Tesla bubble and watch other companies tech presentation.

Shadow mode is a term that Tesla used to describe something known in the industry but yet they never did.
They still today DO NOT do shadow mode according to verygreen. They don't compare what you did with what the computer did. All this have already been debunked.

https://twitter.com/x/status/1096322810694287361

The only one who has copied is Tesla.
Tesla copied their trifocal front camera technology.
Tesla copied their 8 camera setup.
Tesla went "vision only" to copy Mobileye who did it first.
Tesla copied REM to create REM-like maps.
Tesla copied everyone else to do Simulation 1.0 while others like Waymo has already moved on to simulation 2.0
Tesla followed Mobileye to do door to door ADAS after Mobileye revealed it in CES Jan 2020 (L4 as L2+) and announced it with a partner in Sept 2020.

Mobileye, Geely to Offer Most Robust Driver-Assistance Features

www.intc.com

 

[Bladerskb]

I'm not saying Mobileye IS USING Tesla's approach. I'm saying Mobileye seems to be adopting a Tesla-like approach.

As for REM vs Tesla crowdsourcing, see here:

Tesla AI Day 2022

Join us to build the future of AI → https://www.tesla.com/ai--------0:00:00 - Pre-show0:13:56 - Tesla Bot Demo0:29:15 - Tesla Bot Hardware | Hardware Archite...

www.youtube.com

Mobileye obviously went back in time to 2017 to copy Tesla to use REM to generate planet scale realistic simulation road layout/environment.

Gotta love Tesla's fans logic.
Everything that Mobileye did and patented first that Tesla copied years later.
It was actually mobileye who went back in the past and copied Tesla.

https://twitter.com/x/status/1311127306207989765

 

[powertoold]

Mobileye is years behind Tesla. If it's not obvious now, then I'm not the fanboy.

Now they're slowing gaining their fleet, so they're beginning to adopt a Tesla-like approach to developing vision-only fsd.

I'm pointing that out. I'm not wrong here.

Obviously Mobileye isn't going to come out and explicitly say they're adopting a Tesla-like approach. They probably wanted this for a long time now, and it's why they approached Tesla in the first place. But reality is that Tesla is years ahead of Mobileye now. That's why Mobileye is a follower, not a leader, in this approach.

 

[EVNow]

As my Model S85D next summer reaches 8 years and warranty expires I’ve been planning to buy Model Y, but now I noticed that Zeekr plans to launch in Europe this year

If I were you, I'd wait a bit longer and check out Zeekr first and see if it is a good match for you.

Are you thinking Zeekr will offer AV capability in Europe upon introduction. Seems highly unlikely.

 

[EVNow]

Yeah, Mobileye's REM maps are awesome.

I'd wait for 100k cars to be using it all over the continent and getting good reviews before making that comment. But thats just me