Development of Tactical Level AI for Melee and Range Combat

The development of AI directly affects the emergence of new technologies. In modern video games, AI faces a wide range of tasks at various levels. The current situation is such that in addition to standard decision-making, to which the average player is casual, AI often has to do more complex things: to perceive the environment, interact with it, interact with the other AI, move in a complex three-dimensional space and other various tasks. Given the constant development of the gaming industry, the requirements for AI are constantly increasing. Therefore, there is the problem of AI flexibility. In video games, we can increasingly see how the battle of two NPCs turns into a simple search of teams to attack and defend. These primitives repel the player, destroying a decent part of the gameplay conceived by the developer. In the same way it is applicable to the visualization of historical events. For accurate reconstruction, it is necessary that the behavior of the agents be similar to human's behavior. In this paper, we give a brief review on some of well-known AI development methods, compare their effectiveness and present a new method of AI development that simulate the behavior of non- player character in melee and ranged combat based on the interaction of three levels: strategic, tactical and operational for decision-making. Combination of the well-known methods of AI development, base agent's model change and improvement in agent understanding of the environment by using the Voronoi diagram. The method proposed in this paper are showing significantly different results from the most popular design methods and the Utility-AI-Behavior Tree method, significantly reducing the distance in terms of key indicators such as survive time, use of useful resources, number of enemies killed. The used method imitates the player's actions, while excluding the human error factor and unexpected actions. The designed AI simulates the player's logical actions with a good accuracy, but is still more predictable than the real players. Mathematical calculations and the distribution of weights on each frame do not have a significant impact on performance, which allows simulating the behavior of many agents at once in one scenario, without losing performance and influencing the resulting sensations from the gameplay.