Abstract
Among utility-based AI in game AI, there are two types of indicators: goal-based indicators, which have evaluation criteria that guarantee the achievement of the final goal, such as reaching the destination, and non-goal-based indicators, which determine actions based only on local evaluations and do not guarantee the achievement of the final goal. When goal-based and non-goal-based indices are added together and both are taken into account, the problem arises that the goal-based property is lost and the guarantee of goal achievement is lost. The purpose of this research is to realize a behavior that achieves the final goal as quickly as possible, even if both goal-based and non-goal-based are used simultaneously. We propose a method that uses a fill function with a Discrete-Laplacian to maintain the law of transition and guarantee the achievement of the final goal. As an example, we assume that the goal base is the arrival at the destination and the non-goal base is the avoidance of the tracking agents, and the AI that avoids the tracking agents and reaches the destination quickly is realized using the fill function. The adjustment coefficients of the fill function were changed dynamically using fuzzy reasoning based on the numerical values of the AI's fitness and stagnation.
