Eric Bonabeau, François Cogne, Guy Theraulaz

Paper #: 98-01-005

Complex architectures grown by simple agents moving randomly on a 3D lattice and depositing bricks deterministically depending on local configurations of bricks are presented. Some of these architectures are strikingly similar to real wasp nest architectures. But most algorithms, in the space of all possible algorithms of the kind described above, will produce structureless shapes. The space of possible architectures grown by artificial agents on a lattice, and capable of depositing two types of bricks according to the local state of the environment, is huge and long to explore, even if one restricts one's attention to (at least partially) spatially isotropic rules. To overcome this difficulty, a genetic algorithm (GA) has been used with a heuristic fitness function to explore the space of possible architectures. The fitness criterion is based on a simple observation: in algorithms that generate coherent architectures, many micro-rules are used, whereas in algorithms that generate structureless shapes, only one rule or a few rules are actually used in the course of the simulation. Results of this study are reported and discussed.