Genetic algorithms using assembly-instructions as a stand-in for DNA train killbots in an arena.
A dynasty of fighters bred for battle charge the field. The worthy take mates while the losers are cast into the void. But how do we judge a true warriors spirit? By the number notches on his sword? By his deftness of feet? Or possibly whether or not he even took a single bloody step last round, because sometimes that's all we can hope for.
Agents are 255 bytes of 8-bit addressable codespace with genetic algorithms crossbreeding, and mutating simple assembly commands. I've played around a lot with selection schemes and different goals, and while they can develop some rudimentary skills, they always die off and fail to improve. While they can learn a way through a maze, I'm not seeing any emergent combat tactics more advanced than move in a direction flailing my sword around me.
All in all it makes for a fairly retarded AI.
GeneticFighters
An experiment with genetic algorithms and arena combat
Status: Alpha
Brought to you by:
heckruler
Downloads:
0 This Week
Linux