Coverage Path Planning of Agricultural Spraying Drones with Graph Convolutional Network

Optimal route planning for autonomous robots is a crucial element of contemporary agriculture, especially for activities like spraying and pollination. This paper presents a Graph Convolutional Network (GCN) model especially developed to address the Coverage Path Planning (CPP). The aim of this paper is to convert the CPP problem into a Traveling Salesman Problem (TSP), so that by solving the TSP using the proposed neural network, an optimal route can be extracted for each agricultural field in the shortest possible time and with sufficient accuracy, ensuring full coverage of the field's area. The proposed model is a GCN, augmented with a transformer layer and a Multi-Layer Perceptron (MLP), was trained using datasets produced by a Genetic Algorithm (GA) Method. The model obtained a network error of 0.12 while solving TSP cases with 20 points. It showed robust generalization skills, as evidenced by validation and test errors of 0.1261 and 0.1264, respectively. Significantly, the GCN model reduces the solution time to approximately one-tenth compared to conventional approaches, a benefit that increases as the number of points increases. The effectiveness of the model in producing ideal coverage routes highlights its potential for wider use in optimization and coverage path planning, especially in dynamic and extensive agricultural environments.