Research on intelligent content generation and adaptive presentation in virtual simulation scenarios

Abstract

Virtual simulation technologies have become indispensable in education, medical training, psychological therapy, and industrial applications. However, traditional virtual systems rely heavily on static pre-designed content and simple rule-based adaptations, resulting in limited personalization, suboptimal immersion, and increased cybersickness. This paper proposes the Adaptive Generative Virtual Ecosystem (AGVE), a closed-loop framework that integrates intelligent content generation with real-time adaptive presentation. AGVE comprises three synergistic layers: a perception layer using a Conditional Multimodal Variational Autoencoder for uncertainty-aware user state modeling, a generation layer employing a User-Conditioned Latent Diffusion Model for semantically coherent 3D content creation, and an adaptation layer based on Multi-objective Actor-Critic with Online Bayesian Weight Adaptation for dynamic, preference-aware presentation optimization. Extensive experiments in educational virtual chemistry laboratories and therapeutic spider phobia exposure scenarios demonstrate significant improvements: 41.7% relative gain in immersion, 28.3% increase in learning outcomes, 39.9% reduction in cybersickness, and up to 47.8% better task performance. Ablation studies and user feedback further validate the necessity and effectiveness of the integrated architecture. This work advances toward truly intelligent, human-centered virtual ecosystems. The central research question addressed in this paper is: How can a closed-loop framework that tightly integrates real-time user state perception, conditional generative content creation, and multi-objective adaptive presentation simultaneously improve immersion, learning outcomes, task performance, and user comfort in virtual simulation scenarios? The results demonstrate significant improvements across all four dimensions, confirming the effectiveness of the proposed integrated approach.