AutoGrow RL - Adaptive Input Control of an IoT Based Autonomous Greenhouse System for Precision Farming Using Reinforcement Learning

Precision agriculture increasingly relies on autonomous greenhouse systems for optimizing climate control and nutrient management. Our previous work introduced AutoGrow, an IoT-based data sensing and logging system designed to monitor key environmental and soil parameters, including temperature, humidity, moisture, pH, and macro-nutrient (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{N}, \mathbf{P}, \mathbf{K}$</tex>) levels. The system incorporated solenoid valves as actuators for controlled irrigation and nutrient administration, demonstrating the feasibility of AI-driven greenhouse automation. Building on this foundation, we present a simulation study integrating Reinforcement Learning (RL) for adaptive input control, specifically for optimizing macronutrient (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{N}, \mathrm{P}, \mathrm{K}$</tex>) levels and pH balance. We employ Advantage Actor-Critic (A2C) and Proximal Policy Optimization (PPO) to learn optimal nutrient delivery strategies by interacting with a simulated greenhouse environment, leveraging sensor based feedback and dynamic policy updates. Unlike traditional rule based automation, which operates on predefined schedules and static thresholds, this study explores the potential of adaptive, data-driven nutrient management. Simulation results indicate improvements in nutrient use efficiency and system adaptability, suggesting plausible applicability in real world autonomous greenhouse systems.