Simulating Polyculture Farming to Learn Automation Policies for Plant Diversity and Precision Irrigation

Polyculture farming, where multiple crop species are grown simultaneously, has potential to reduce pesticide and water usage while improving the utilization of soil nutrients. However, it is much harder automate polyculture than monoculture. To facilitate research, we present AlphaGardenSim, a fast, first order, open-access farming simulator with single plant growth irrigation models tuned using real world measurements. AlphaGardenSim can be used for policy learning as simulates inter-plant dynamics, including light competition between plants in close proximity approximates greenhouse garden at 25, $000\times $ speed natural growth. This paper extends earlier work new action space that includes planting, which dynamically finds seed locations increases resources utilization, an adaptive sampling technique number actions taken each timestep without affecting performance. We also evaluate other automation policies novel metric combines diversity canopy coverage. Code supplementary material found https://github.com/BerkeleyAutomation/AlphaGarden . Note Practitioners —Monoculture often characterized by heavy agrichemical inputs, such chemical fertilizers pesticides, increased vulnerability disease pestilence. motivated lack long-term sustainability industrial agriculture, its implications human food security. Although sustainable alternative monoculture requires more labor challenging automate. In this propose order setting. Simulation experiments suggest learn watering pruning plan robot follow produce maximal yield from diverse set limited irrigation, however not yet been tested on physical garden. future research will develop fully automated controller operate tools over cycles.