Sensor-Based Scouting Algorithms for Automatic Sampling Tasks in Rough and Large Unstructured Agricultural Fields

Abstract

Labor shortage has prompted researchers to develop scouting robots to facilitate agriculture field sampling tasks. The purpose of this study was to develop and evaluate sensor-based automatic scouting algorithms for planning efficient paths to guide scouting robots over all sampling points in rough unstructured agricultural environments. A triangular mesh map was used to represent the rough agricultural field surface, with the map incrementally built in simulations. The adapted line-sweeping scouting algorithm was composed of two parts: a coverage algorithm that identified a reasonable coverage path to traverse sampling points, while a dynamic path-planning algorithm determined an optimal path between two adjacent sampling points. Sample points were defined a priori, given a desired spatial sampling density, and the coverage path list was defined as those sample points in sequence of the line-sweeping direction. The coverage task was completed when all accessible sampling points had been traversed. The dynamic path-planning algorithm was used to find the optimal path between adjacent sampling points, circumventing any obstacle between them. The sensor-based scouting algorithm was able to cope with different situations with or without an a priori map or in a partly known environment. The performance of this algorithm has also been evaluated by comparing with two other methods via simulations. Results showed that the path length, energy requirements, and time requirements of the adapted line-sweeping method were less than those for the potential function method and the bug method for four different physical starting points within a single agricultural field. Further, the number of scans required using the adapted line-sweeping method was less than those required for the bug method.