MobilEye's approach

[lunitiks]

Isn't Mobileye = Intel now? Like *unlimited* legal resources?

So if - if - there is a legal showdown, it will surely be huge. Like totally massive. If.

(Just my 2 decapitated presidents' heirs)

 

[Bladerskb]

The video you show doesn't actually show the camera placement (where it mounts on the car), only the angles. However, the angles are different on the Tesla. The forward facing side cameras cover the side view for Tesla, while the rear facing side cameras cover the side view for Mobileye.

same placement, same angle

 

[croman]

Isn't Mobileye = Intel now? Like *unlimited* legal resources?

So if - if - there is a legal showdown, it will surely be huge. Like totally massive. If.

(Just my 2 decapitated presidents' heirs)

You can't assume that, due to conflicts, Intel's outside counsel will be the same as ME. Or anything really about how their decision-making is structured. Maybe Intel will disclose that as part of their SEC filings but I don't believe that level of detail is known. I could be wrong.

 

[stopcrazypp]

those are just some of their patents.



same angle
let me post a video.

That video's from November 2016 though and is a different angle than the one shown in the 2015 video. Tesla can easily claim prior art.

I didn't miss anything. that lawsuit deals strictly with trade secrets that was stolen by a former employee and has nothing to do with actual patents. If you actually followed any part of that lawsuit, you will know this. That lawsuit will also evolve to be a criminal investigation.

The article says specifically they are asserting claims on 4 patents.

 

[Bladerskb]

That video's from November 2016 though and is a different angle than the one shown in the 2015 video. Tesla can easily claim prior art.

The article says specifically they are asserting claims on 4 patents.

They never once talked about patent throughout the lawsuit so far everything that has been public is related to trade secret. As uber has already said they are not using any of Google patent as they are using an off the self velodyne lidar on their cars. Everything has been focused on the trade secret.

In fact just now

Judge William Alsup, who is presiding over the civil lawsuit, referred the case to the U.S. Attorney tonight — a move that could result in a criminal investigation into Levandowski’s behavior.

“This case is referred to the United States Attorney for investigation of possible theft of trade secrets based on the evidentiary record supplied thus far concerning plaintiff Waymo LLC’s claims for trade secret misappropriation. Said evidentiary record is described in detail in the Court’s order, also issued today, on Waymo’s motion for provisional relief. The Court takes no position on whether a prosecution is or is not warranted, a decision entirely up to the United States Attorney,” Judge Alsup wrote in the referral.

Waymo’s claims of trade secret theft could result in criminal case

 

[gowthamn]

Mobileye made/makes great technology. But when watching Amnon always remember he has a lifetime of research to defend - and an approach to autonomous driving to justify as well. The unsupervised approach was not possible even 36 months ago because the GPU power simply didn't exist to run these networks. Mobileye's supervised learning, compartmentalized approach to vision was the only way to execute for the last number of years. Unsupervised end-to-end is a threat to Mobileye's business model.

So mobileye does not use Deep reinforcement learning?

 

[gowthamn]

From where is Tesla getting HD maps?

Does Mobileye use Deep Learning?Bit confused.

 

[Snuffysasa]

Mobileye made/makes great technology. But when watching Amnon always remember he has a lifetime of research to defend - and an approach to autonomous driving to justify as well. The unsupervised approach was not possible even 36 months ago because the GPU power simply didn't exist to run these networks. Mobileye's supervised learning, compartmentalized approach to vision was the only way to execute for the last number of years. Unsupervised end-to-end is a threat to Mobileye's business model.

What are you talking about?

what unsupervised approach?? Unsupervised learning is not being used in any real autonomous system that I know of. And neither is end-to-end learning. (image->steering angle)

 

[Snuffysasa]

So mobileye does not use Deep reinforcement learning?

Mobileye is doing research on deep reinforcement learning that will run on deep recurrent neural networks, that will run on Intel silicon. This may not happen for several years though.

 

[Bladerskb]

Mobileye is doing research on deep reinforcement learning that will run on deep recurrent neural networks, that will run on Intel silicon. This may not happen for several years though.

mobileye already uses deep learning in their eyeq3 chip

 

[gowthamn]

Mobileye is doing research on deep reinforcement learning that will run on deep recurrent neural networks, that will run on Intel silicon. This may not happen for several years though.

Ok, but Tesla uses CNN for vision and Deep Reinforcement learning for driving policy?

 

[lunitiks]

Ok, but Tesla uses CNN for vision and Deep Reinforcement learning for driving policy?

CNN? I'll give you CNN

 

[gowthamn]

CNN? I'll give you CNN

CNN is Convolution Neural Network.

 

[Snuffysasa]

mobileye already uses deep learning in their eyeq3 chip

Agreed. they are using deep learning, but not deep reinforcement learning.... well I wouldn't say the eyeQ3 is doing any kind of "learning." Simply using a trained CNN for inferencing. The eyeq4 and eyeq5 will use much more CNNs and bigger ones.

However, neural nets that mobileye trains with deep reinforcement learning, which will most likely not be a CNN. Those will not run on EyeQ and will have to run on a separate host processor. But this is a couple years away.

Ok, but Tesla uses CNN for vision and Deep Reinforcement learning for driving policy?

Yes Tesla uses CNN for Tesla vision.

Who knows what Tesla is doing for driving policy or if they are even considering driving policy yet.

Driving Policy is not necessary for L4 fully autonomous driving. Driving Policy will just make future L4 L5 cars more efficient and better at handling complex traffic scenarios. But like I said this is years away.

It is possible that Tesla autopilot will use some form of neural net for path planning / decision making / control for FSD. but there is no details about this or nothing to confirm this that I am aware of. I wouldn't be surprised if no neural nets / AI is used in Tesla cars other than for Vision. Also there is no details or evidence that Tesla is doing any deep reinforcement learning either.

 

[Bladerskb]

Agreed. they are using deep learning, but not deep reinforcement learning.... well I wouldn't say the eyeQ3 is doing any kind of "learning." Simply using a trained CNN for inferencing. The eyeq4 and eyeq5 will use much more CNNs and bigger ones.

However, neural nets that mobileye trains with deep reinforcement learning, which will most likely not be a CNN. Those will not run on EyeQ and will have to run on a separate host processor. But this is a couple years away.

Sorry i most have skipped over the question and missed the 'reinforcement' part. Mobileye however has created and continues to create driving policies for their eyeq4 that is based on reinforcement learning.

Driving Policy is not necessary for L4 fully autonomous driving. Driving Policy will just make future L4 L5 cars more efficient and better at handling complex traffic scenarios. But like I said this is years away.

However i disagree with you that driving policies aren't needed for L4(and i assume you meant something like highway, etc).
The example that Mobileye put out is a good reason why.

 

[Snuffysasa]

Sorry i most have skipped over the question and missed the 'reinforcement' part. Mobileye however has created and continues to create driving policies for their eyeq4 that is based on reinforcement learning.


However i disagree with you that driving policies aren't needed for L4(and i assume you meant something like highway, etc).
The example that Mobileye put out is a good reason why.

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.


And I did not mean something like highway. Driving policy is not needed for a fully autonomous city driving.

 

[Bladerskb]

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.


And I did not mean something like highway. Driving policy is not needed for a fully autonomous city driving.

source? because mine disagrees strongly.

 

[Snuffysasa]

source? because mine disagrees strongly.

how does that disagree?
No where does it say that EyeQ3 or 4 uses something related to deep reinforcement learning.

 

[gowthamn]

Driving Policy is not necessary for L4 fully autonomous driving.

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?

 

[Bladerskb]

how does that disagree?
No where does it say that EyeQ3 or 4 uses something related to deep reinforcement learning.

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