# AgroNova: An Autonomous IoT Platform for Greenhouse Climate Control

## TL;DR

AgroNova is an autonomous IoT system for greenhouse climate control that uses sensors, local decision-making, and a language model to manage the environment efficiently.

## Contribution

The study introduces a hybrid IoT architecture combining edge autonomy with context-assisted reasoning, validated in a real greenhouse.

## Key findings

- The system achieved an average local reaction latency of less than 1 second during threshold exceedance events.
- Physical actuator operations occurred in only 2.3% of control decisions, indicating a conservative control strategy.
- The system operated continuously even during internet outages due to autonomous gateway-level control.

## Abstract

This study presents AgroNova—a hybrid IoT architecture for autonomous monitoring and management of microclimate in greenhouse environments. The system combines a capillary wireless sensor network, gateway-level rule-based agents, a server agent, cloud services and an advisory component based on a large language model (LLM) that supports local decision-making by incorporating external contextual information from meteorological services. The proposed architecture was validated through a seven-month deployment in an unheated tomato greenhouse, during which more than 380,000 environmental measurements were collected from five sensor nodes. The system operated continuously under real agricultural conditions, including during temporary internet connectivity interruptions, due to the autonomous gateway-level control and deterministic fallback mechanisms. The analysis of the collected data includes 3110 environmental threshold exceedance events, in which recovery dynamics, reaction latency, and actuator activation frequency were evaluated. The results show that the architecture supports stable autonomous operation under limited actuation conditions, with an average local reaction latency of less than 1 s, while physical actuator operations occur in approximately 2.3% of all control decisions. This behavior reflects a conservative control strategy that limits unnecessary mechanical operations and contributes to stable system operation. The experimental integration of a consultative LLM module within the server-side agent demonstrates the potential for context-enriched decision support using external meteorological data, while final control decisions remain under the authority of the gateway-based deterministic control mechanism. The main contribution of this study is the demonstration of a hybrid IoT architecture that combines edge-level autonomy with context-assisted reasoning, validated through deployment in a real greenhouse environment.