Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
I level 4 decisions regarding lane merges, to yield or not, may need a policy.

Anyway, so currently once CNN derect something the decision making is all hard coded / rule based?

sure in that sense you need a policy of course. But when mobileye talks about driving policy, they are referring too negotiation with human drivers. "the rules of breaking the rules"

and yes autonomous cars use CNNs for perception then use rule based decision making. correct.

driving policy algorithms runs on eyeq4, it says it right there...

but I said deep reinforcement learning.

Me: "Nothing in mobileye eyeq3 or eyeq4 is the product of deep reinforcement learning.

Mobileye is developing algorithms created with deep reinforcement learning, and these algorithms will run on other host processors not EyeQ."


Also, even with driving policy. this chart just means, that mobileye EyeQ4s will be used in autonomous cars that have driving policy, but not that driving policy algorithms are running on the Eyeqs themselves
 
sure in that sense you need a policy of course. But when mobileye talks about driving policy, they are referring too negotiation with human drivers. "the rules of breaking the rules"

and yes autonomous cars use CNNs for perception then use rule based decision making. correct.



but I said deep reinforcement learning.

Me: "Nothing in mobileye eyeq3 or eyeq4 is the product of deep reinforcement learning.

Mobileye is developing algorithms created with deep reinforcement learning, and these algorithms will run on other host processors not EyeQ."


Also, even with driving policy. this chart just means, that mobileye EyeQ4s will be used in autonomous cars that have driving policy, but not that driving policy algorithms are running on the Eyeqs themselves

nope, everything listed there are algorithms and tech that mobileye has in their chip. however OEM can create more driving policies using the template that mobileye created which they can run on the chip. All driving policies that mobileye creates is also available to any OEM on that eyeq4 chip if they decide to use it. Mobileye also told us the cost of running these policies on their chip.

Yes Driving policy are created with deep reinforcement learning but these models WILL run live on the eyeq4 chips.
 
nope, everything listed there are algorithms and tech that mobileye has in their chip. however OEM can create more driving policies using the template that mobileye created which they can run on the chip. All driving policies that mobileye creates is also available to any OEM on that eyeq4 chip if they decide to use it. Mobileye also told us the cost of running these policies on their chip.

Yes Driving policy are created with deep reinforcement learning but these models WILL run live on the eyeq4 chips.

Source??

Mobileye CTO says the driving policy algorithms created with deep reinforcement learning will run on Intel silicon.
 
That's what I was going to post.
Watch the video closely, in both instances he says the deep reinforcement learning algorithms will run on Intel Silicon.

that's for their partnership with intel/bmw and delphi.
Bmw uses intel and mobileye chip, including delphi.

But for other OEM who don't have a partnership with intel can run those policies on the eyeq4 itself and they can use multiple eyeq4.

As he says "this can run inside our eyeq4 without us even feeling it"
 
@Snuffysasa

You asked me how i interpreted Mobileye chart.
Well they said they have five l3 production and five l4 production programs.

I also know that BMW, Nissan, NIO and lucid motor all have agreed to REM.

Audi, Honda and Volvo on the other hand have yet to public-ally agree.

We also publically know that Audi L3 is coming out in 2017, Nissan L3 is coming out in 2018 and BMW L3 in 2021. That leaves Volvo and Honda for 2019 launch


B4F9Rz1.png
 
that's for their partnership with intel/bmw and delphi.
Bmw uses intel and mobileye chip, including delphi.

But for other OEM who don't have a partnership with intel can run those policies on the eyeq4 itself and they can use multiple eyeq4.

As he says "this can run inside our eyeq4 without us even feeling it"

Why would those partnerships run on intel silicon if they can just run on other eyeq4s. Also I think multiple eyeq4s is just to add in more and more cameras and sensors.. You cannot reprogram an EyeQ to do certain tasks, they are all the same and are black boxes.

""OEM who don't have a partnership with intel can run those policies on the eyeq4 itself "" source?
 
@Snuffysasa

You asked me how i interpreted Mobileye chart.
Well they said they have five l3 production and five l4 production programs.

I also know that BMW, Nissan, NIO and lucid motor all have agreed to REM.

Audi, Honda and Volvo on the other hand have yet to public-ally agree.

We also publically know that Audi L3 is coming out in 2017, Nissan L3 is coming out in 2018 and BMW L3 in 2021. That leaves Volvo and Honda for 2019 launch


B4F9Rz1.png


So Volkswagen signed on with REM so I assumed that meant Audi too, but idk.

Yea I guess I agree with your conclusion about Volvo and Honda in L3. maybe they are yet to be publicly announced. And also its possible this is out of data and road maps have changed.

I wasn't sure how to read the chart like, there is a start and end of the gray bar,. what does the start point mean? and what does the end point mean?

And for the L4 chart: are we assuming Lucid and Nio in 2019. And Audi and Nissan in 2020?
 
Why would those partnerships run on intel silicon if they can just run on other eyeq4s.

Because the Intel SOC handles the control so they use it for driving policy aswell.
Same way, Audi's zFAS is a SOC that also handles control, if audi were to use driving policy from mobileye, they will run it on their zFAS central computer.

Also I think multiple eyeq4s is just to add in more and more cameras and sensors.. You cannot reprogram an EyeQ to do certain tasks, they are all the same and are black boxes.

I believe eyeq3 handles actual algorithm control for ADAS and doesn't just output its vision.

""OEM who don't have a partnership with intel can run those policies on the eyeq4 itself "" source?

Ammon literally says they have run it on their eyeq4 and it takes less than 1% of eyeq4 computing.
He also said "this can run inside our eyeq4 without us even feeling it"

Goto 41 mins 20s on that video.
 
Because the Intel SOC handles the control so they use it for driving policy aswell.
Same way, Audi's zFAS is a SOC that also handles control, if audi were to use driving policy from mobileye, they will run it on their zFAS central computer.





Ammon literally says they have run it on their eyeq4 and it takes less than 1% of eyeq4 computing.
He also said "this can run inside our eyeq4 without us even feeling it"

Goto 41 mins 20s on that video.

"I believe eyeq3 handles actual algorithm control for ADAS and doesn't just output its vision."
Hmm I don't know that it does, I know EyeQ2 definitely does not, pretty sure EyeQ3 doesn't.

Hmm and you are right he does say that it runs on EyeQ4 here... I am corrected for now. thanks for pointing this out. I will look into it carefully in the morning.
 
So Volkswagen signed on with REM so I assumed that meant Audi too, but idk.

Yea I guess I agree with your conclusion about Volvo and Honda in L3. maybe they are yet to be publicly announced. And also its possible this is out of data and road maps have changed.

I wasn't sure how to read the chart like, there is a start and end of the gray bar,. what does the start point mean? and what does the end point mean?

And for the L4 chart: are we assuming Lucid and Nio in 2019. And Audi and Nissan in 2020?

Lucid first car is coming out in 2019 and NIO already announced a L4 autonomy 2019 date.

NIO - NIO and Mobileye Joining Forces to Deliver L4 Autonomy by 2019

idk about start/end of the gray bar. I feel like who ever made that chart was overwhelmingly drunk

Also Volvo came out in April, 3 months after this CES talk and said they were skipping level 3.

So everything is still kinda fluid.
 
Lucid first car is coming out in 2019 and NIO already announced a L4 autonomy 2019 date.

NIO - NIO and Mobileye Joining Forces to Deliver L4 Autonomy by 2019

idk about start/end of the gray bar. I feel like who ever made that chart was overwhelmingly drunk

Also Volvo came out in April, 3 months after this CES talk and said they were skipping level 3.

So everything is still kinda fluid.

Yea I know Nio was delayed to 2020.

And so if Volvo is skipping L3. Then that makes me very confused about this chart. Why Volvo was even on it.
 
Yea I know Nio was delayed to 2020.

And so if Volvo is skipping L3. Then that makes me very confused about this chart. Why Volvo was even on it.

well this talk was done Jan 2017 and Volvo came out against L3 april 2017.
so maybe they changed their mind.

“In this mode the car is in charge of the driving, yet the driver must still be prepared to take over in case of emergency, which could be a matter of a few seconds. Volvo considers this Level 3 driving mode unsafe and will thus skip this level of autonomous driving,” Mr Samuelsson said.

Volvo Cars CEO urges governments and car industry to share safety-related traffic data

Although they have had previous negative comments about L3, they were probably on the fence and had plans for both L3 and L4.
 
Last edited:
Yet another extremely informative talk by Amnon Shashua, really worth watching. He even mentions Tesla a couple of times:

Interesting, but his argument against the "map heavy" approach of Waymo is unconvincing. Mobileye, like Tesla, would have nothing to sell to consumers today if they developed a Lidar based autonomous driving system. Waymo doesn't need to sell car automation to consumers/auto manufacturers today.

Waymo is playing the long game. Both Tesla and Mobileye are making claims about autonomy and falling short on delivering results.

What mobileye is selling to auto manufacturer's today is a possible path to not be dominated by Waymo-based autonomous cars. GM and others are pursuing multiple development paths (Cruise Automation and Mobileye) to maximize the possibility of a good outcome. Tesla could have reduce risk by keeping the Mobileye relationship while doing its own development.
 
  • Disagree
Reactions: lunitiks
I think that was the most convincing part, actually

To paraphrase Shashua from the video, "The most difficult part of heavy mapping is the expense". The big auto companies will not spend a couple billion dollars driving vehicles around to laser map the world. But what about google? Consider "heavy mapping" in google's broad corporate strategy.

Shashua also talks like he has superior AI capability compared to Google. Yet google is running a beta of level 5 in Phoenix while Mobileye has a power point presentation. Meanwhile Tesla is trying to match the capability of Mobileye's old tech.
 
  • Funny
  • Like
Reactions: croman and lunitiks
To paraphrase Shashua from the video, "The most difficult part of heavy mapping is the exp The big auto companies will not spend a couple billion dollars driving vehicles around to laser map the world. But what about google? Consider "heavy mapping" in google's broad corporate strategy.

Exactly but if they did want it then they can use HERE HD maps that uses lidar. GM for example hired a company to do map the entire highway system (160,000 miles) with lidar.

the new mercedes drivepilot 4.5 uses here maps