Abstract
In this paper, we introduce a skill-balancing mechanism for adversarialnon-player characters (NPCs), called Skilled Experience Catalogue (SEC). Theobjective of this mechanism is to approximately match the skill level of an NPCto an opponent in real-time. We test the technique in the context of aFirst-Person Shooter (FPS) game. Specifically, the technique adjusts areinforcement learning NPC's proficiency with a weapon based on its currentperformance against an opponent. Firstly, a catalogue of experience, in theform of stored learning policies, is built up by playing a series of traininggames. Once the NPC has been sufficiently trained, the catalogue acts as atimeline of experience with incremental knowledge milestones in the form ofstored learning policies. If the NPC is performing poorly, it can jump to alater stage in the learning timeline to be equipped with more informeddecision-making. Likewise, if it is performing significantly better than theopponent, it will jump to an earlier stage. The NPC continues to learn inreal-time using reinforcement learning but its policy is adjusted, as required,by loading the most suitable milestones for the current circumstances.