Agbots: Weeding a field with a team of autonomous robots

Authors: MCALLISTER, WAYNE - UNIVERSITY OF ILLINOIS; OSIPYCHEV, DENIS - UNIVERSITY OF ILLINOIS; DAVIS, ADAM; CHOWDHARY, GIRISH - UNIVERSITY OF ILLINOIS

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/18/2019
Publication Date: 8/1/2019
Citation: McAllister, W., Osipychev, D., Davis, A.S., Chowdhary, G. 2019. Agbots: Weeding a field with a team of autonomous robots. Computers and Electronics in Agriculture. 163. Article 104827. https://doi.org/10.1016/j.compag.2019.05.036.
DOI: https://doi.org/10.1016/j.compag.2019.05.036

Interpretive Summary: Cost-effective weeding robots are urgently needed to cope with multiple herbicide resistance in weeds of field crops, but none are yet up to the task. Patchiness of weed populations makes random or row-by-row navigation (typical of current models) clearly not the most efficient way to manage weeds with a limited robot workforce. Instead, a radically different approach is needed: coordinated deployment of robots based on real-time information sharing and worker recruitment. We simulated robotic weeding of a field with a patchy weed population. Seven simulation experiments were run, varying how much information was shared among the robots, number of robots, robot speed, and seedbank population density. Information sharing among robots improved weeding performance, as did robot working speed. The number of robots needed increased linearly with seedbank population density, indicating that the problem did not become unsolvable as weed seedbanks grew larger. Overall, the results support an approach to robotic weeding that uses coordinated teams of robots, operating autonomously but sharing information about field conditions.

Technical Abstract: Cost-effective weeding robots are urgently needed to cope with multiple herbicide resistance in weeds of field crops, but none are yet up to the task. Newly emerged weed population densities in a field can vary by orders of magnitude. This high spatiotemporal dependence makes random or row-by-row navigation (typical of current models) clearly not the most efficient way to manage weeds with a limited robot workforce. Instead, a radically different approach is needed: coordinated deployment of robots based on real-time information sharing and worker recruitment. Foraging, where robots move through an environment with the goal of collecting objects or information, has long been considered a key problem in multi-agent robotics. In our case, the foraging problem is framed in terms of recognizing and killing weeds while moving through the field. Using a dynamic programming algorithm, we simulated multi-agent robotic weeding of a field with a spatially heterogeneous weed seedbank, with pulses of weed seedling recruitment over time. Seven simulation experiments were run, varying the degree of information sharing among the agents, number of agents, agent speed, and seedbank population density. Information sharing among agents improved weeding performance considerably, as did agent velocity. The number of agents required scaled approximately linearly with seedbank population density, indicating that the problem did not become increasingly complex as weed seedbanks grew larger. Overall, the results represent a proof of concept for an approach to robotic weeding making use of coordinated teams of robotic agents, operating autonomously but sharing information about field conditions.