Towards Electromagnetic-Field Physical AI for 6G: Concepts, Architecture, and Standardization

Le résumé fourni par la source

Physical AI is conventionally associated with robots, vehicles, and other mechanically embodied systems. This article extends Physical AI to electromagnetic-field embodiment, where RF-programmable hardware acts on real electromagnetic fields and propagation environments via phased arrays, reconfigurable intelligent surfaces, and programmable wireless environments. We define electromagnetic-field Physical AI through a perception–modeling–decision–execution–feedback loop, introduce a four-level hierarchy from passive perception to closed-loop field-state regulation, and propose five identification criteria to distinguish field-level intelligence from conventional parameter optimization. We further position this perspective relative to emerging world-model-based wireless intelligence, arguing that the two are complementary: world models emphasize predictive internal cognition, whereas electromagnetic-field Physical AI emphasizes physical-layer embodiment through which intelligent policies act on propagation environments. We examine the control-theoretic structure of distributed-parameter field dynamics, discuss reduced-order modeling and structured physical priors, and analyze implications for 6G architecture and standardization. A concept-validation example suggests that explicit field-level optimization can outperform parameter-centric baselines under simplified but explicit controlled assumptions. This work also proposes a layered standardization framework with field-level intent descriptors and capability abstractions, offering a starting point for future programmable wireless environment standards.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.