A dog-training game idea

I know of at least two 'games' that are similar in scope and idea, these are "Serious Games" in that they're used by particular groups to estimate and evaluate the movement of individuals and to develop tactics that humans might not have thought of yet.

Simplistically speaking; agents within the game are 'trained' that it is a bad thing to get killed and a good thing to kill others, so through a number of iterations of simulation they develop tactics that engage each other that has red beat green, or green beats red, they put the AI against humans to see if human factors defeat the tactics of the AI, and when they're satisfied that the AI tactics could work in reality... they extract the tactics and teach these approaches to humans and put them in the same scenarios in-in the field training environments.

The point? Training animals in the simulation that you're describing isn't too far from the sims that i've seen, meaning from that perspective that its a reasonable concept (its been done before, for example), wasn't immediately abandoned (is in use today), and has applications ongoing (who knows you might break through with a K-9 Bot that is more than just a jumping puppy).

I would definitely prototype something, but recognize that AI is not a simple topic and will be tough, there are a number of books on the topic of developing AI, and Neural Networks are just one way of doing machine learning and has lots of jargon associated to it, lots of interesting math, and some mindbogglingly strange implementations from a casual observer perspective.

Sorry i just thought i'd add...

I'm sure that there's still people out there today playing those old CATZ and DOGZ games, and would probably sell well (if you got to that point) in the Japanese market and other areas where ownership of animals is limited...

@Stragen

Thanks for the response. I guess I'm mostly looking for a short fun game to develop, instead of a serious big one. I've dabbled in neural networks previously, and if that's anything to go by, I'm afraid half of all the work will be figuring out the structure and tuning. Which... yeah, I'll be frank, leaves little time for the fun parts. But if dog reactions directly correspond to your commands, that kind of takes the whole motivation away. I'm hoping maybe I'll be able to fake it in some way.

CATZ and DOGZ are virtual pet games, right? I have never even heard of them. Maybe I'll get inspired. I mean, it should be clear that my main focus is that there be artistically represented dogs.

Getting the boring bits done is probably going to be the lions share of the work, absolutely, but isnt that really the way of things?

The CATZ, DOGZ, were basically the PC versions of the tamagochi and really the fore runner to the 'play with a virtual pet' style things, kinda like the groom your Pikachu aspects of the new Pokemon game.

Surprisingly the PETZ (of which CATZ and DOGZ was a part) series was developed by Ubisoft (amongst others) and had all sorts of things in them, but nothing neural net AI, more state machines IIRC, and simple Pavlovs Dog (pun intended) approach to interacting with the animal; pat dog = happy, kinda stuff. I think another older one that claimed to do some more advanced AI was "Creatures" but then i'm not sure how that worked under the hood.

i'd be keen to see what you come up with if you do go ahead and prototype