Hi all I'm trying to implement simple multi-agent artificial life system, like prey and predator. At this moment, i think, that agents will be very simple creatures, controlled by artificial brain (preferably by neural network). So, questions is about: 1. Network architecture: multilayer, self-organization networks, reccurent networks, ... 2. Learning methodics and criteria 3. How agent will estimate environment? Separate environment into sections (far, near, left, right, back, front, ...), give each section weight (for example, negative means danger, positive food) and make complex analysis of environment, or estimate each neighbor agent separately and than apply rule "winner takes all"? 4. Desicions choosing. Agent must move to get to the food and get away from predators, eat plants to maximize energy. Thanks for all your answers in any way. ps: i found following videos very interesting:

Quote: Original post by chak
At this moment, i think, that agents will be very simple creatures, controlled by artificial brain (preferably by neural network).

Why?

Quote: Why?

It is possible to implement fake agent AI using simple techniques, like state-machines and another. But, it is interesting to make brain, that can learn and improve "thinking quality" :) And, neural networks usage is one of the items in technical requirements.

[Edited by - chak on January 29, 2009 4:35:21 AM]

Quote: Original post by chak

Quote: Why?

It is possible to implement fake agent AI using simple techniques, like state-machines and another. But, it is interesting to make brain, that can learn and improve "thinking quality" :) And, neural networks usage is one of the items in technical requirements.

I'll say it once again, then I'll drop the subject. Articifical Neural Networks have nothing to do with the brain, just like Non-Linear Programming isnt about coding. If you go using ANN because you think you will make a "brain" that can "think" and "learn", you're headed for a very bad place. Still, good luck.

[Edited by - Steadtler on January 29, 2009 2:27:36 PM]

lol \=

I heard about ann lightly more than nothing. And, have done some tiny projects:
- objects classification (for a given set of characteristics indicate object membership in a particular class)
- contour construction from b\w and colour images
- pid controller with neural network for non-minimum-phase systems

Now i'm just interested, how can i use ann in agents ai.