Make NPC More Realistic: Design and Practice of a Hybrid Stealth Game NPC AI Framework Based on OODA Theory

Stealth game is a game genre that mainly focus on avoiding the enemy’s detection, in which the intelligent behavior of NPC is crucial to create an excellent player experience. This requires not only vivid and diverse NPC appearance, actions, and dialogue, but also real and credible interactions in game. However, to develop a stealth game NPC AI system is difficult, and it often shows some shortcomings in terms of diversity, autonomy, and emotionality. This paper introduces a hybrid NPC AI framework for stealth games based on OODA (Observe-Orient-Decide-Act) theory. Unlike the Observe-Decide-Act cycle used by most stealth games, our framework incorporates an “Orient” element, enhancing NPC believability in narrative and gameplay by emphasizing unique character traits. Our framework consists of four layers. The Observe layer integrates visual and auditory perception at the sensory level, generates distinct stimuli events for customized stimulus-response relationships. The Orient layer incorporates emotional, factional, identity, and reaction pattern for personalized decision-making. The decision layer supports dynamic environment triggers and scripted events. The Action layer interacts with the game world, and it also adds vivid dialogue, voice, and animation to enhance player immersion. The designed framework is implemented in a third-person, dual-protagonist, action-stealth game developed under Unreal Engine. Players control one protagonist and command the other, navigating dangerous levels filled with menacing enemies to silently approach and eliminate targets. This paper plans to use comparative experiments and questionnaires to validate the effectiveness of this framework in enhancing player experience.