Implementation of a deterministic video game agent testing environment

Competitive multiplayer video games take many forms. From short, slow-paced, strategic card games, to fast and realistic first-person shooters, all have one thing in common – the presence of agents controlled by an artificial intelligence. The human factor often requires the application designer to perform various tweaks, as human behavioral patterns are complicated and evolve based on the individual game knowledge. Adaptive artificial agents are undoubtedly the best solution, but call for prolonged testing. The article describes ways to implement determinism in an AI testing environment, vastly decreasing the required number of individual agent training sessions. In the analyzed scenario, testers participate in a setnumber of matches, facing only human opponents in the environment which keeps track of their movement and actions. Then an artificially controlled agent is introduced into the recorded sessions, and once again the statistics of each participant are analyzed to determine whether the AI can adequately adapt. In the case of failure, AI parameters are changed, and the process is repeated until the result is acceptable for the game designer. The environment has been designed to test a novel algorithm that has been proposed in the previous work and allows to change agent pathfinding to maximize the entities score (i.e., the number of kills), depending on current and previous player actions.