DustNET: Enabling machine learning and AI models of dusty plasmas

Dusty plasmas, ubiquitous throughout the universe, may be found in laboratory and industrial plasmas, nuclear fusion devices, Earth's atmosphere and ionosphere, the lunar surface, planetary rings, cometary tails, and the interstellar and intergalactic media. Despite decades of experimental, theoretical, and computational research, many aspects of dusty plasmas remain poorly understood within a unified predictive framework, i.e., the Standard Model for dusty plasmas. Existing first-principles models based on particle, kinetic, fluid, and compound descriptions have successfully explained numerous phenomena, including dust charging, transport, waves, and self-organization. However, predictions remain challenging because of the inherent multiscale and multiphysics nature of dusty plasmas; computational complexity and other computing limitations; and incomplete knowledge of material properties, transport coefficients, and boundary conditions. Recent advances in machine learning (ML) and artificial intelligence (AI) provide new opportunities to complement traditional physics and first-principles modeling through data-driven scientific inference. This paper surveys current ML and AI applications in dusty plasma research, including microparticle imaging and tracking, image analysis, surrogate modeling, physics-informed learning, geometric learning, uncertainty quantification, and multimodal data fusion. We then introduce Dust Neural nEtworks Technology (DustNET), a community-oriented data infrastructure and benchmark ecosystem designed to support curated dusty plasma datasets spanning laboratory experiments, industrial plasmas, planetary, space, and astrophysical observations, numerical simulations, and AI-driven methods. We further discuss the Dusty-plasma Unified Sensing, Testing and Multi-modal Analysis Platform (DUST-MAP), a proposed member of dusty plasma world models that learn unified representations from heterogeneous dusty plasma data. By integrating experiments, measurements, simulations, scientific data and metadata, physical constraints and other regularizations within a common representation-learning framework, DUST-MAP aims to support universal inference and predictions across heterogeneous instruments, experimental facilities, and natural dusty plasmas. Toward the end, we present a phased community roadmap for the co-evolution of DustNET and DUST-MAP, emphasizing reproducible benchmarking, AI agents, and edge intelligence toward an AI-enabled ecosystem for dusty plasma science, with scalable and data-driven scientific intelligence across laboratory, industrial, planetary, space, and astrophysical dusty plasma environments.