SOFTWARE ARCHITECTURE AND IMPLEMENTATION OF A DATA-DRIVEN ADAPTIVE AІ SYSTEM FOR 2D TACTICAL GAMES

This article addresses the problem of limited replayability and the predictable behavior of non-player characters (NPCs) in modern 2D tactical simulation video games. The study examines the design and software implementation of a flexible, adaptive artificial intelligence (AI) system that operates without relying on asymmetric game balancing techniques («cheating»). The objective of the research is to develop the architecture and functional modules of a game capable of dynamically responding to a player's strategy in real time. To construct the decision-making system, the Hierarchical Task Network (HTN) planning method was employed, integrated with utility functions for action evaluation and selection. Spatial and tactical analysis of the game environment was implemented using Influence Maps and the A* graph-based shortest pathfinding algorithm. The software implementation of the project was carried out using the modern game engine Godot Engine 4.5, following a data-driven approach through hierarchical JSON configuration files. A modular architecture was developed, comprising the tactical analyzer, terrain manager, and config loader subsystems, which enables the computation of complex HTN plans within a background thread. This approach ensures high computational efficiency, supports realistic group behavior of enemy units, including flanking maneuvers, tactical retreats, and position holding, and enhances the overall depth and strategic complexity of the gameplay experience.