Openpilot
- Anmelden oder Registrieren um Kommentare zu schreiben
George Hotz, the hacker who jailbroke the iPhone first at 17, is now making an "open-source" ai for driving. I saw this through a invidious video
https://invidio.us/watch?v=Nnh5TQ60hek
I found the github for the self-driving ai at
https://github.com/commaai/openpilot
at the very bottom of the github page there seems to be an MIT license
https://www.gnu.org/licenses/license-list.html#X11License
this seemed closest to what I think it is? Not sure if this is free or not. I am very interested though! Let me know your thoughts, if it were free software would you use it?
it seems that they are selling the hardware without software (for legal purposes), and that is the reason why the software is open-sourced. Something phone-sized goes on the dash of the vehicle and also has cameras facing towards the driver to see if they are paying attention to the road.
The more I think about it, the only concern might be tivoization where the free software is locked behind the hardware that you need to purchase from their company, but if the code is truly free then it should be able to be adapted to a different device. What prevents a competitor.. or freeloader from simply forking the code?
Self Driving Cars sound like a really, really bad idea. Especially given google's captcha is being used to train their algorithms to run self driving cars. My point I guess? I refuse to buy one of those or use one if I can help it.
I don't know if google is involved with all self driving cars, but even so, it could be dangerous. Who knows what backdoors are within them. :/
If it were free software, backdoors could be identified through the source. I don't think that it is a self-driving car yet, it NEEDS human attention still.
Not sure if this is free or not.
Grepping the source code for a few minutes, I found no license notice that would state some code is nonfree. That does not mean the code is all free software: it is huge and I have not searched for things like binary disguised as source code.
There are actually many incompatible licenses in the source tree: the GNU GPL, the MPL, the CDDL, the Eclipse license, etc. Those incompatible licenses appear in different sub-directories of either the "external" or the "phonelibs" directory. It may well be that the related pieces of code are never linked (in which case, it is fine).
The more I think about it, the only concern might be tivoization
Is the hardware tivoized? A reference is missing.
zapper wrote:
Who knows what backdoors are within them.
You do not, obviously. Unless you point at a blob in the source tree or explain where the backdoor is in the free code, it is just one more groundless accusation from you.
>Is the hardware tivoized? A reference is missing.
I was asking the question, not trying to say it is.
>Grepping the source code for a few minutes, I found no license notice that would state some code is nonfree. That does not mean the code is all free software: it is huge and I have not searched for things like binary disguised as source code.
Would an audit of such a large amount of code be possible? How are binary blobs hidden within open-source programs detected? Would it be possible for me to do it with a program, on an x200, even if it takes a few hours?
Would an audit of such a large amount of code be possible?
According to 'sloccount', 73% of the 420k lines of code are in the "phonelibs" directory and 12% in "external". Most of that would be dependencies if packaged for a GNU/Linux distribution. For instance, 110k lines of code are in phonelibs/eigen (for linear algebra, the "libeigen3-dev" package in Trisquel's repository), 77k lines in phonelibs/acado (a toolkit for for Automatic Control And Dynamic Optimization) and 32k in external/cppad (for diferenciation, eponymous package in Trisquel's repository).
The new critical code looks concentrated in the "selfdrive" directory, which contains fewer than 55k lines of code. It is certainly possible to audit that code for safety and security (I believe it is what you mean by auditing). Notice that section "Safety and Testing" of https://raw.githubusercontent.com/commaai/openpilot/master/README.md suggests the developers take the problem seriously. Also, having dependencies (rather than reinventing the wheel, but with bugs) is good practice.
How are binary blobs hidden within open-source programs detected?
The program could rely on nonfree binaries. Nevertheless, as I wrote, I have not found any explicitly nonfree license in the source tree. When it comes to detecting binary disguised as source code, the Linux-libre project executes https://www.linux-libre.fsfla.org/pub/linux-libre/releases/LATEST-5.N/deblob-check (but it is kind of specifically tailored to discovering blobs in Linux's source code).
Notice that I do not see any good reason to doubt that openpilot is free software. It does not hurt to investigate anyway.
>Notice that I do not see any good reason to doubt that openpilot is free software. It does not hurt to investigate anyway.
Exactly what I was thinking. When I saw that a company was making an "open-source" ai I thought that it was too good to be true, that is why I was trying to see whether it is free or not.
> When it comes to detecting binary disguised as source code, the Linux-libre project executes https://www.linux-libre.fsfla.org/pub/linux-libre/releases/LATEST-5.N/deblob-check (but it is kind of specifically tailored to discovering blobs in Linux's source code).
I had seen a post on these forums where jxself was talking about deblobbing the linux kernel himself, which led me to ask whether deblobbing could be done on this program. Of course it might not be necessary.
>The new critical code looks concentrated in the "selfdrive" directory, which contains fewer than 55k lines of code. It is certainly possible to audit that code for safety and security (I believe it is what you mean by auditing).
When you have software take control of a car, not only does the code have to be working flawlessly, but yes I did mean an audit from someone independent from the company creating it. So that anything that might record data (cameras,gps,any sensor) could not theoretically 'phone home' to the company.
Thanks for the answers!
> if it were free software would you use it?
No.
Even if it was completely isolated from the internet, it would be a temporal fun then I would get tired of it soon. Because I like driving. If experienced once with someone's car, it would be enough.
I think it has no value other than (re)highlighting some already known issues. I am not sure if it is worth to spend a lot of effort just highlighting the issues and taking risk.
I am a member!
I do not like cars in general. They are a huge burden on the poor and
the disabled, and they are a waste of money for the middle class.
I very much enjoyed cars in Europe. It’s actually fun driving through autobahns. In the States it’s the mode of transportation that is the first and last resort. Since it’s more of a luxury in Europe, I felt completely fine never getting in a car except the occasional taxi. I fell out of love with cars after driving here in the US.
I think that it would be a great leap for free software if such a complex ai were to be made truly free. I do not yet know enough about this project. Worst case scenario it could be a great teaching tool for artificial intelligence as a whole. Maybe nothing will come of it, who knows.
> They are a huge burden on the poor and
the disabled,
I might agree. The sharing system seems to be quite well developped in our country, though.
> and they are a waste of money for the middle class.
How?
> Worst case scenario it could be a great teaching tool for artificial intelligence as a whole.
Software developers might be able to compete for kind of AI GT cars races. It would not require huge money unlike Formular 1. AI cars races would be fun to watch regardless of classes of cars from F1 to Prius than installing the system on my car. I think those cars may be faster than human.
The AI would drive very better than most people though. It might be safer for me. I only have a scooter. Nasty drivers are not that rare.
I am a member!
On 3/2/20 12:49 AM, name at domain wrote:
>> They are a huge burden on the poor and
> the disabled,
>
> I might agree. The sharing system seems to be quite well developped in
> our country, though.
What do you mean by 'sharing system'? If you mean Uber or Lyft, these
are not good, because they require proprietary software.
In the U.S., there are very few transportation services. You MUST own
and operate a car to get anywhere. This means disabled people have to
stay at home and poor people have to go into debt to get to work.
>> and they are a waste of money for the middle class.
>
> How?
Insurance and upkeep costs on cars is very expensive. If we had a
decent bus system, we wouldn't be under so much pressure to spend all
that money.
--
Caleb Herbert
KE0VVT
(816) 892-9669
https://bluehome.net/csh
Cars/traffic kill every two years more people than were killed in action in Vietnam 1955–1975 (American casualties).
> Uber or Lyft
Yes.
> they require proprietary software.
That's why I mind to use ths system. If I could secure anonimaty, I would want to use them occasionally, though.
> disabled people have to
stay at home and poor people have to go into debt to get to work.
Buses and trains are like sardine cans in some areas of Japan. But surely the decent bus system seems to be a good or maybe the best solution at present. Hope double pay and half working time will solve all this kind of issues in the future.
I did not know that that is a that serious issue in the U.S.. I had heard of that vaguly though, like the U.S. is large, so everyone needs a car.
> Cars/traffic kill every two years more people than were killed in action in Vietnam 1955–1975 (American casualties).
I have been thinking that a traffic accident is the current best way to fill thanatos. I would not say that every driver would aim for it, it is a sensitive issue, though. Some of them would have commited the crime "unintentionally".
In Japan, in the lightest case, we can commit murder by paying less than 1,000,000 yen fine or less than 7 years in prison.
- Anmelden oder Registrieren um Kommentare zu schreiben