Hacker Built a Self-Driving Car In His Garage - Bloomberg

[jimmy82abn]

Yes you are correct in this blog but his first version was using Iphone cameras. It's just interesting to see someone who could be the leader of the autonomous world in the next 5 years. He has already been offered a job by Elon. Just depends what he wants to do with his version of autopilot.

“Frankly, I think you should just work at Tesla. I’m happy to work out a multimillion-dollar bonus with a longer time horizon that pays out as soon as we discontinue Mobileye.”
​

Hotz replied:

“I appreciate the offer. but like I’ve said, I’m not looking for a job. I’ll ping you when I crush Mobileye.”
​

Musk answered with his well-known one-word email “OK”.

 

[Benz]

George Hotz built a self-driving car. In his garage. Competition for Mobileye?

Bloomberg: http://www.bloomberg.com/features/2015-george-hotz-self-driving-car/

The Verge: http://www.theverge.com/2015/12/17/10374422/elon-musk-george-hotz-self-driving-car-tech-criticism

 

[SebastianR]

Bloomberg: http://www.bloomberg.com/features/2015-george-hotz-self-driving-car/

The Verge: http://www.theverge.com/2015/12/17/10374422/elon-musk-george-hotz-self-driving-car-tech-criticism

We discussed this multiple times (over in the News section). Bottom line as far as I see it: there are problems that need one silver bullet. There are problems that need a lot of lead bullets. Self-driving cars needs tons of lead bullets (data, machine learning) to be reliable in all the edge cases.

This is not a problem of "one brilliant mind working for 5 years" this is more a problem of "a team of 2000 really good guys need to work for 2 years".

Ergo: nope, Geohotz - no matter how brilliant he is, will not be a serious thread to MobileEye.

 

[bonnie]

Hacker Built a Self-Driving Car In His Garage - Bloomberg
Interesting Article
Auto Pilot Recreation with a Smartphone

And also discussed in numerous threads not focused on this particular topic (a lot of the AP ones).

 

[Ulmo]

We discussed this multiple times (over in the News section). Bottom line as far as I see it: there are problems that need one silver bullet. There are problems that need a lot of lead bullets. Self-driving cars needs tons of lead bullets (data, machine learning) to be reliable in all the edge cases.

This is not a problem of "one brilliant mind working for 5 years" this is more a problem of "a team of 2000 really good guys need to work for 2 years".

Ergo: nope, Geohotz - no matter how brilliant he is, will not be a serious thread to MobileEye.

I appreciate your discussion, so please don't take this personally. I see these points brought up time and time again, and they're valuable for what they are.

I find it ever so slightly amusing to see both the Big Data and Little Data prejudices against higher quality AI efforts. I think all three are needed: high quality AI (better and more brains), high quality sensors (better and more sensors), and high quality data (better and more data). At a certain point, shooting for all three gets you there so much faster that you kind of over-shoot, but right now, over-shooting would be way better than what we currently have. Getting that San Francisco guy + Google Big Data + massive sensor arrays of all types will clearly solve this quickly, but will require quite a bit of money, cooperation with the competition, getting over your pride about not having to do the stuff the other guy does, etc.. For now, we're watching the efforts in this field just sort of cobble around in their own little corners. A few of them will figure it out eventually, regardless of how, but this need to claim that you HAVE to do it a certain way when in fact that's probably completely wrong is just normal corporate cattle calling as far as I'm concerned.

 

[Jeff N]

Unmentioned in the articles I've seen about Holtz is that the major Silicon Valley company Nvidia is following a similar approach using multi-layered neural network AI technology along with custom chips designed to accelerate the real-time processing. I attended a session at CES in Las Vegas recently where they showed impressive progress and disclosed a new generation of hardware chips that they claim will fit into a "lunch box" sized package and only cost a few hundred dollars.

They have put together a developer platform and are providing a default pre-trained neural network database based on real world driving and image detection. They hope to sell their solution to car companies for automated driving. The website is here:

Autonomous Car Development Platform | NVIDIA DRIVE PX|NVIDIA

 

[xmetal]

Looks like Mr. Hotz is continuing his work and getting closer to product reality. Saw this video today and it looks like comma.ai is starting to be a "thing." Interesting to see how this stuff will pan out in the near future.

 

[Cosmacelf]

OK, it doesn't slow down for turns. There was a bug that it didn't stop for a stopped car. He had to take over to not take a freeway exit.

BUT, this looks really promising. I hope they can pull it off.

 

[Electroman]

He is relying on machine self learning through crowd sourcing, which is what I think it will take eventually to make this more autonomous on all conditions of the roads. Something like, Waze, for traffic information. He gave an example of how his car learnt to handle those white bots as lane markers almost overnight, whereas Mobile eye had to spend years and millions to get that in, and still not good as theirs.

Don't like his seeming arrogance and teenage like gestures

 

[HankLloydRight]

Don't like his seeming arrogance and teenage like gestures

He's like a young Quentin Tarantino.

 

[bhzmark]

Looks like Mr. Hotz is continuing his work and getting closer to product reality. Saw this video today and it looks like comma.ai is starting to be a "thing." Interesting to see how this stuff will pan out in the near future.

Very interesting. An odd bird, clearly on the spectrum and arrogant and immature and many other signs of genius and likelihood to be a productive disruptive force if the immaturity is held in check.

Neural net machine learning is clearly the way to go for something like this and the Vegas dots is a good example of that. Also very clever to sell a dash cam as a way to collect that data.

I hope tesla has already been collecting this data from their million miles a day drivers and is in an even better position to feed the crowdsourced data to a neural net to improve on current driving software.