An ART-Bev-Based Cognitive-Behavioral Framework for Personality-Driven Simulated Agents

Abstract

Modern multi-agent collaborative simulations demand more realistic, adaptive, and interpretable computer-generated agents (CGA). Existing models often exhibit rigid behaviors and weak individual differentiation. To address this, we propose a personality-driven cognitive framework that integrates the Big Five model, a task-oriented need hierarchy, valence-based affect modulation, fatigue accumulation, and an interpretable ART-Bev decision-rule learner. The framework realizes a five-layer closed loop spanning cognitive attributes, situation awareness, decision reasoning, behavior execution, and effectiveness evaluation. Its methodological novelty lies in explicitly coupling stable traits, dynamic needs, affect, and fatigue into a compact decision cycle, then compiling learned rules into auditable behavior trees. We evaluate the framework on NetLogo in a multi-agent terrain game. Cross-terrain tests, parameter sensitivity analysis, ablation studies, and dual-expert interpretability validation show consistent gains in task success, terrain transfer robustness, behavioral diversity, interpretability, and runtime efficiency over conventional baselines. The results support the use of personality-driven CGA in complex cooperative and competitive simulation settings.