Towards autonomous greenhouse crop production: virtual environment development, AI training and configuration

Summary

This thesis aimed to develop an AI solution for autonomous greenhouse climate control and crop management, which is made up of a simulation model, multiple AI agents and a set of configuration management mechanisms for adapting the AI solution to different real environments. Operation of the AI solution follows the three steps of the route map. First, the simulation model is configured towards the target real greenhouse regarding greenhouse structure, equipment and crop cultivar. Then, AI agents for greenhouse climate control and crop management are trained in the simulated environment with reinforcement learning algorithms. Finally, the AI agents are migrated to the real greenhouse, for which the observations of the real environment are configured to reduce the effects of uncertainties before they are sent to the AI agents. This thesis developed the AI solution with a case study of greenhouse tomato production.