Closed-loop LLM framework autonomously optimizes plant growth and energy use in vertical farming

A new arXiv preprint introduces a closed-loop LLM framework that processes data from a 49-channel phytosensor network—covering multispectral, electrochemical, and dielectric modalities—to autonomously control hardware actuators in vertical farming setups.

The system transitions from human-in-the-loop analysis to autonomous control by evaluating plant physiology in real time and triggering actuators to optimize microclimates, execute phenotyping protocols, or induce controlled stress scenarios.

Validated across three case studies, including a vertical farm and a single-plant setup, the framework deciphered complex micro- and macro-fluctuations in plant physiology and executed multi-parameter optimization balancing biomass accumulation, chlorophyll content, and energy consumption.

In a production-scale deployment, the LLM modulated full-spectrum, 450 nm, and 660 nm lighting at 2-hour intervals. Compared to periodic control, the system reduced the production cycle by 35% in minimal-time mode and cut energy consumption by 18% in energy-optimization mode with only a marginal increase in cultivation time.

Agents autonomously developed an unforeseen strategy of dark-induced chlorophyll accumulation, resulting in a 67.9% energy saving while maintaining cultivation outcomes.