Predictive Robotic Harvesting Based on Non-Stationary Markov Chain Vegetation Cycle Model and Crop Simulation

Abstract

This article provides methodological foundations for the predictive harvesting of berry fruits with autonomous robots. By predictive harvesting we mean here a procedure to assign the robots tasks in the areas of the crop that guarantee the best yield or the satisfaction of other indicators of harvest efficiency. The reported research focuses on the construction and subsequent validation of a vegetation cycle model of strawberry in covered crops such as foil tunnels and greenhouses. The model is calibrated with real-life observations of selected strawberry variants growth that depends on insolation, temperature, and humidity. The vegetation cycle model is embedded into crop simulation that records all fruits picked by robots in the greenhouse. The latter application calculates sites with the highest expected ripe fruit density to start the next day harvest. The multicriteria optimization procedure maximizes the harvest yield and harvest efficiency, and minimizes the mass of non-harvested ripe fruits.