That's more or less what I'm thinking. Part of the problem comes in splitting it up into managable chunks to actually get somewhere.
What I mean is that the finished product, if we tried to code it all at once, could easily take months. I'd definately like to split it into managable mini projects that each take maybe a few weeks to get working.
The only downside is that it really eliminates the possibility for balancing explicitly. Like the relative cost for shell and slime, things like that, because those properties then become emergent.
Metabolic DNA is going to have an interesting structure I think. I'm not sure the best way to go about it. The problem can be thought of like this:
Given a series of vats, each containing various volumes of water, and hoses connecting those vats, find the most cost effective way to achieve a desired volume level in certain vats.
Should the program automatically decide how best to distribute the vats given the bot's desired levels? Or should the bot have to figure out for itself exactly how to redistribute the vats?