RtNEAT AI engine resource

#1 Is there a website or something that tells us a bit more about rtNEAT?

#2 Yes! I'm very interesting about this resource.

Steven

#3 Here is more info on the rtNEAT project

http://nn.cs.utexas.edu/downloads/papers/stanley.cig05.pdf

I am just dealing with Licensing issues with the University of Texas

The Trusted One

#4 @Steven,

do you know anything about AI programming and neural networks, I could definitly use a wing man on this one.

#5 Cool!
I've been looking through the abstracts on utexas and it seems very interesting.

What are the licenses for?
Do they have their own library or are there pattents involved?

#6 Being not familiar with neural networks I am nevertheless very interested!

Any ideas what a license will cost?

#7 The source has been released GPL - Non-Commercial, so if the pack is built and released free(which is the plan) then I am we are NOT in violation of the exsisting license.

The problem comes in with the fact that even an indie game is inheirently a commercial venture. I have contacted the powers that be at U of Texas plus the coder who wrote it, nothing to do but wait for a response

I have quoted the email I sent U of Texas below if anyone is curious.

Quote:Rick,

I am aware of the rtNEAT projects history with the Torque game engine. What I am proposing to do is to create a resource for their engine that would allow for the large number of Torque developers to use, and possible evolve upon this technology.

If you will pardon me I must take a moment and acknowledge the incredible work of your students, and thank you for your help in allowing gifted developers such as Ken to push our industry forward.

With that aside here are the paticulars of what I am proposing to do.

I wish to build an add-on pack for both the Torque Game Engine and the Torque Shader Engine, that will integrate Ken's work into the engine. This add-on I will release to the community at no cost, I will do this because like the Garage Games community I believe in helping the independant developers grow and evolve. Now with that said, we have the conflict that the base license for the engine is GPL Non-Commercial. For my base idea of the module being released at no cost I dont see a conflict, however, despite the fact that most of the projects being built with the Torque engine are being done by "indie" developers with nothing that even resembles a budget, they are, for the most part, at there nature a commercial venture.

I must admitt it has been along time since any of my teams projects have had anything that resembles a budget.

I have had in mind either a nominal one time licensing fee per developer, upon release of their product. Paid to you directly by the developer. Less preferably possibly a royalty based payment system once a project that utilizes this technology has reached a certain number of sales.

I do ask that when considering this you take into account the type of licensing that Garage Games themselves implements, for more information on that visit the website at http://www.garagegames.com.

I am hoping that we can come to an agreement that would allow the independant video game community to benefit from and I truely hope help evolve this technology.

Thank you for your time and consideration,

Michael Kowalski
Digital Awakenings
Software Engineer

#8 Ok, thanks for clearing that out Mike :-)

Why not ask for a license change to LGPL?
I that way the product could be use any anyone commercial or non-commercial
as long as the source for any changes to the rtNEAR library is published.
That's definitely a solution I would choose (both as a IP owner or a IP licensee).

#9 Because I am sure the University want to reserve the right to charge companies who have the resources through the nose for the license. This actually works in our favour as it give us an advantage over the billion dollar giants, and those are hard to come by. Also the rtNEAT was proven using torque to create the NERO game

The Trusted One

#10 If you're sure your' re sure :-)

I still think LGPL is more beneficial for all, but that's maybe just me being an "Open Source fantastic" ;-)

One argument I would point out is that "the billion dollar" giants wouldn't have much trouble hiring
someone(s) to write it from scratch. After all its all been officially published in papers?
(Or am I missing something vitally important and unique about the rtNEAR?)


But if your sure UT isn't interested in a LGPL release, there's not much to argue about :-)

Let us know how the negotiations turn out, I love to hear what the conditions and
the price of a license would be.
(hopefully much less then the cost of doing something from scratch?)

Cheers,
Johan

#11 Has anyone compile the rtNEAT C++ package under VC2003 or higher? I created and VC project and compiled this last week and got some obscure namespace related errors and haven't revisited it again. Was interested in running the XOR problem which I am very familiar with from prior work. It's sort of my starting point problem for testing a NN.

#12 I briefly looked at the code, It doesnt appear to be designed to stand alone. I did not explore to deeply yet I may be WAY off base on that.

Johan: I will most certainly try to go for the full open source, I might be able to get them to issue that license for Torque owners. I will have to see how it goes.

The Trusted One

#13 Mike: Sounds good :-)

Tim: I don't have any luck compile rtNEAR 1.0 with VC2003 either. Lucks like some work is needed and my VC/C++ knowledge isn't expert in anyway :-(

#14 Very interested in this ressource.

#15 Tim:

Ok I got rtNEAT compiled with VC2003.

Here's my setup:
I've started a new Console project and added everything but the nero_evolution.cpp (which is just a demo of integration in a GameEngine).

In Project Properties:
I had to turn of "Create/Use Precompiled Header" which was on by default (in the C/C++ compile options for the project).

In experiments.h:
I moved the const static from the private part of the class in to global scope. (VC2003 complains about const static doubles as class members. Don't know why?)

In neatmain.cpp:
The time() function needed is header includes (e.g. added #include ).

In experiments.h:
The go_cart() function uses lrand48() for some calculation. I don't know how to replace it properly (I'm not a UNIX not VC2003 C++ expert as I might have mentioned :-)
Anyway I just replaced it with rand() and hoping that the modulus that follows will ensure that a sensible value is used.

Any feedback on my bas assumptions are welcome :-)

Regards,
Johan

#16 Question about integration of rtNEAR in Torque (from a relatively newbee C++ programmer).

As I've understood it Torque doesn't play well with STL (so I've avoided it so far).

But rtNEAR uses the SDT::vector template for its data structures.

Would this be a problem with Torque?
Should it maybe be changed to use core/tVector instead?

#17 Johan: Thanks for the headsup, that will save everyone using VC++ a little time.

#18 Pleasures all on my side :-)

Beware that I still get a lot of warnings and it might be necessary to do some more work.
(But the XOR example seems to finish correctly :-)

#19 Day dreaming a bit about this kind of software.
(Hopefully not to much bs :-)

The NEAR software seems to have some interesting features (I'm not sure if they are in rtNEAR,
but I'm guessing the could be migrated quite easily).
One thing is the time delay feature of the neurons (which hasn't been use by UT in any published research).

I think using this feature would make the network a spiking neural network, it already have some spiking
features in the rtNEAR version because signals are passed on to the next layer of the network every sim tick.

With the time delay, this stimuli to response would be even more sequential,
and my hypnoses is that a SNN would result in a more "planing" kind of problem
solving. Where different sequential steps could be learned.

The result of NERO for instance seems more to be of a direct response kind of intelligence, e.g.
not appearing very cunning.

Being able to learn sequences of responses would give the agents more of a plane ahead kind
of response.

Well that's just a hypnosis, that I hope going to be able to prove in the future :-)

Another interesting aspect that could be explored, imho, is inter-agent communication.
E.g. learning the agents to talk to each other.

I envision Jurassic-park/Aliens kind of games here!
E.g. avoiding being eaten by dinosaurs or "aliens"
in the jungle by sneaking around listening to the creatures "talking" to each other. The creatures listening
for sound I make as I move through the AI-controled vegetation.
The AI of the vegetation is for calculating sound and visual "effects" based on my
movement-behaviour. Who said NN should only simulate brains? :-)

Well, you can say I'm really exciting about this :-)

#20

Quote:Now with that said, we have the conflict that the base license for the engine is GPL Non-Commercial.

Where did you reached the conclusion that the license has anything to do with the GPL? From what I can see it's simply a custom license that prevents "commercial use" but grants you the right to make use of the source in your own software so long as it meets their definition of "non commercial".

This is far from been a GPL compatible license in that it for starters prevents commercial use which the GPL allows, also it has no copyleft style clauses. So really its a "custom non-commercial".

A plus with it not been GPL related is that many of the Torque based projects on here are non-commercial so for those projects the source could be used with Torque without issues, unlike if it was GPL'd.

That said, I'd be interested in hearing anything you find out about the cost of a commercial license. Not that I have a use for this yet, but I'm still interested in knowing :)