Animats and Decision Making

Conceptually, the tactical decision process is about mapping features of the game situation and criteria for the decision onto the desired behavior: FxC=B; where C=[c₁, c₂, K, c_m] is the set of criteria, and B=[b₁, b₂, K, b_p] represent the possible behaviors.

Just as with the weapons, it's possible to assess the fitness of a behavior, given the features and criteria: FxCxBR. Indirectly, this allows the best behavior to be extracted, because we can compare the fitness of each behavior.

To determine these values in practice, various methods can be used. Deduction is used to combine fixed knowledge about strategies (for instance, ambushes are likely in a valley), whereas experience would learn from examples (for instance, this spot is vulnerable). These processes along with hybrid combinations are discussed further in Chapter 22.

Monitoring the performance of tactical behaviors is the major problem. Behaviors are executed over periods of time, so the criteria apply continuously. Therefore, the benefit of the behavior needs to be accumulated over time, converting incremental feedback into a fitness value for that period.

The animat's decisions also need to take into account the fact that behaviors will be executed over time, and their benefits may not be immediate.