Desenvolvimento de comportamento adaptativo de NPCS em ambientes dinâmicos gerados proceduralmente

Abstract

This work presents the development of a procedural generation system for indoor environments and an adaptive behavior model for an NPC, applied to an exploration game set in a haunted mansion. The map generation process employs Binary Space Partitioning (BSP), graphs, and post-processing to create rooms, connections, and navigable routes. The ghost’s behavior was implemented through a finite state machine integrated with sensors and the A-star algorithm for path planning. Experiments with 100,000 maps were conducted by varying dimensions and parameters to analyze structural connectivity. The results showed that the rate of fully connected maps strongly depends on the generation parameters, ranging from approximately 80% to 96% across the tested scenarios. The performance of A-star heuristics was also evaluated, with average times very close to each other: Manhattan produced the shortest execution time, while Chebyshev and Euclidean distances showed differences below 3%. The results indicate that the proposed approach generates coherent and varied maps, as well as an NPC with reactive and contextually understandable behavior. It is concluded that the integration of procedural generation and behavioral AI is effective for creating dynamic indoor environments, providing a solid basis for future extensions.