From Agentic AI to Physical AI: World Model-Empowered Internet of Embodied Agents for 6G Networks

Abstract

As 6G evolves toward integrated communication, sensing, computing, and intelligence, Physical AI is transforming digital agents into embodied agents capable of perceiving, reasoning, and acting in the physical world. However, existing Internet of Agents (IoA) frameworks provide limited support for embodied task orchestration, reasoning–control coordination, and future-state prediction. To address these challenges, this paper proposes a World-Model-Enabled Internet of Embodied Agents (IoEA) architecture. First, a ground-based global Digital Agent coordinates multiple Unmanned Aerial Vehicles (UAVs) through global perception, task orchestration, and dynamic scheduling. Second, each UAV hosts a distributed Embodied Agent with a hierarchical Vision-Language-Action (VLA)/Actor mechanism that converts high-level task semantics into high frequency continuous actions, enabling real-time execution with reduced dependence on centralized models. Third, a Latent World Model (LWM) predicts environmental evolution, shifting multi-UAV collaboration from reactive to predictive coordination. The proposed architecture enables the synergistic integration of digital cognition and physical action in 6G networks.