BrilliANT: An Entry for the GECCO’2007 Ant Wars Contest

One of our first observations concerning Ant Wars was that ant’s field of view (FOV) is rather small (24 fields when excluding the actual ant’s position). If one considers food locations only and takes into account rotational invariance and symmetry, there are 2/4/2 = 2 = 2097152 unique FOV states (when ignoring the constraint placed on the amount of food and the existence of opponent). As FOV occupies approx. 20.7% of the ant’s world and the total number of food pieces amounts to 15, the expected number of food pieces within FOV is slightly more than 3 when the game begins. Also, the probability of having n food pieces within FOV drops quickly as n increases and, for instance, for n = 8 it amounts to less than 0.5%. This further reduces the number of realistically possible FOV states by several orders of magnitude. This suggests that it is difficult to build a sophisticated strategy based only on the current FOV state. Probably more may be gained by virtually extending the FOV, i.e., keeping track of board state as the ant moves. To enable this, we equip our ants with memory, which involves the following three components, each of them implemented as a two-dimensional array overlayed over the board: