Large Language Models for the Internet of Things

The Internet of Things (IoT) has evolved from isolated sensor deployments into large-scale, heterogeneous cyber-physical ecosystems spanning industrial plants, smart cities, mobility, energy, agriculture, and healthcare. Yet IoT systems still face persistent barriers: fragmented device and data semantics, high operational complexity, brittle automation logic, and a widening gap between raw telemetry and actionable decisions. Meanwhile, Large Language Models (LLMs) have demonstrated strong capabilities for reasoning over natural language, code, and multi-modal inputs, enabling planning, tool use, and interactive decision support.A compelling next step is to treat LLMs as semantic middleware and reasoning engines for IoT bridging human intent and machine execution, translating unstructured operational knowledge into executable policies, and providing context-aware automation across devices and services. This shift enables higher-level interactions such as “explain why this alarm is recurring,” “generate a mitigation workflow,” or “optimize energy under comfort constraints,” while also supporting machine-to-machine coordination through structured tool calling and agentic orchestration.However, IoT introduces constraints and risks that make naive LLM adoption impractical: edge resource limits, intermittent connectivity, real-time requirements, privacy and regulatory constraints, safety-critical actuation, and adversarial threat models. As with other Frontiers Research Topics, this collection should unify theory-to-practice contributions connecting architectures, methods, datasets, and deployments so the community can move from prototypes to robust, measurable IoT+LLM systems.This Research Topic aims to advance the scientific and engineering foundations for integrating LLMs with IoT systems, with emphasis on: (i) semantic understanding of IoT data and operations, (ii) reliable planning and orchestration across devices and services, and (iii) trustworthy, privacy-preserving, and safe deployment under real-world constraints. We seek contributions that provide validated methods, benchmarks, and reproducible implementations that clarify what works, when, and why.We welcome manuscripts that propose, analyze, or empirically validate IoT+LLM approaches across the stack (device/edge/cloud), including but not limited to these sub-themes:- Semantic IoT and interoperability: LLM-assisted ontology alignment, schema mapping, intent-to-policy compilation, natural-language interfaces to digital twins.- Edge LLMs and resource-efficient intelligence: small/quantized models, on-device inference, split computing, latency/energy trade-offs.- IoT data grounding for LLMs: retrieval-augmented generation (RAG) over time-series, logs, and knowledge graphs; sensor-to-text and multimodal representations; uncertainty calibration.- Autonomous orchestration and agents: tool calling for device control, workflow synthesis, planning under constraints, human-in-the-loop oversight.- Trust, safety, and security: prompt injection in IoT contexts, secure actuation, access control, privacy-preserving learning, red-teaming, guardrails, auditability, and compliance.- Benchmarks and deployments: datasets, evaluation protocols, reproducibility packages, and field studies in industrial IoT, smart buildings, mobility, and healthcare. Topic editors: Heena Rathore, Shailendra Rathore Source: https://www.frontiersin.org/research-topics/78414/large-language-models-for-the-internet-of-things