For my final year project for university I was given the title "Toy Planet" with the brief that I was to create some sort of a-life simulator. My tutor suggested I just build a straight up genetic algorithm with a population of animals and veg, and measure the tendency for particular schemata (alleles, basically) to evolve.
I thought that idea was boring, so I built a simulator capable of simulating agents with "dna". The dna was a string of characters of any length which acted on a simple 10-item stack, performing basic conditional jumps (jump if greater than, jump if not zero, short jump etc.), setting a state (veg, eat, mate) and moving up/down/left/right. I allowed the mutation rate of individual agents to be mutated itself, and sought to prove that evolved mutation rate is a function of dna-repair costs and sustained biodiversity. In the paper itself, I actually messed that up, but after I'd submitted it I figured out why, and actually managed to get exactly that result
Anyways, in a lot of ways it's similar to Darwinbots, which is why I bring it up. Not only will you guys probably be interested in playing with it, but also I specifically designed it in a modular fashion so that huge changes to the processing of DNA, physics etc would be very easy.
What this means is that there currently exists a framework which can very easily be modified to incorporate the best features of Darwinbots, with much cleaner code, and native cross-platform support (it's in Java). Also, the networking stuff is already set up because the university let me run it on their distributed computing rig :3
I can't release it yet because I have to wait for marking to finish, but there's some info and a live sim that you can view at
http://www.ameliapollard.co.uk/EvoSim/.
Any thoughts on this as a potential candidate for DB3?