As an independant study I am planning on doing a project in the field of neural networks. I have formulated an idea, but would like to know what others think (is it possible, potential snags, etc) Build two neural networks that play a game of "tag". One is the persuer (predator) the other the evador (prey). The neural networks would work off eachother and train in tandom. The input would most likey a simple distance between the two objects involved in the game. The network would then output a vector as to where to head next. Do you think this is enough information (maybe the input would also need orientation)? I plan on just starting the two objects on an endless open plane. If it works, I would like to train them on a plane with obstacles and see if I can get them to avoid walls as well as use them to their own advantages. So... do-able? Thanks!

yup

http://www.cogs.susx.ac.uk/users/davec/pe.html

You may also find the tutorials at my website useful.

---

ai-junkie.com

Thanks fup!

Nice link.

Cool, a link to my uni, better yet, a link to the work im copying (i mean errm using as a starting point for my project).
Keep us updated on the progress cos im really getting into all this pursuit-evasion coevolutionary stuff.

quote:

---

Original post by fup
yup

http://www.cogs.susx.ac.uk/users/davec/pe.html

You may also find the tutorials at my website useful.

---

ai-junkie.com

---