NPCRadar: NPC-Centered VR Multi-User Crowd Forecasting

Multi-user virtual reality (VR) mechanisms enable immersive entertainment experiences in which multiple users share a common virtual environment and continuously influence one another through their movements and interactions. These evolving crowd states have a direct impact on how non-player characters (NPCs) should be responded to. However, existing research on VR behavior understanding mainly concentrates on modeling behavior at the individual level. To address this limitation, we introduce the task of NPC-centered crowd forecasting in multi-user VR, which aims to predict short-horizon future crowd states in relation to a target NPC. We characterize crowd states using three complementary dimensions, such as Flow, Density, and Intent, providing a structured representation of group-level dynamics around the NPC. Building on this formulation, we propose NPCRadar, a framework that establishes an end-to-end pipeline from historical multi-user behavioral observations to group-level future state prediction. By extending VR behavior understanding from individual-level analysis to crowd-level forecasting, our study offers a principled foundation for anticipatory NPC control and facilitates more intelligent, natural, and socially aware interactions in multi-user VR environments.