Розробка адаптивного штучного інтелекту NPC для тактичних симуляторів з використанням методу GOAP

Abstract

This work presents the development of an adaptive artificial intelligence system for non-player characters (NPC) in tactical simulators using the Goal-Oriented Action Planning (GOAP) method. Unlike traditional approaches based on finite-state machines (FSMs) or behavior trees, the proposed system enables agents to dynamically build action sequences in response to changing environmental conditions. The architecture is implemented in the Unity game engine using C# and includes a perception module, a GOAP planner based on A* search in the state-action space, and a library of atomic actions. The system demonstrates emergent tactical behavior, including flanking maneuvers, suppression fire, cover usage, and coordinated group actions in close-quarters battle (CQB) scenarios without hard-coded scripting