In mountainous orchards, agricultural tasks, such as crop protection and harvesting, are characterized as being labor intensive and dangerous. An autonomous orchard robot that can execute these unattended seems a promising alternative to increase task operability. An essential function in the development of an autonomous orchard robot is navigation, which is usually based on tree-row detection from LIDAR scan data by using navigational algorithms. This research applies a probabilistic particle filter (PF) algorithm with a novel laser-beam model for the autonomous in-row navigation of an orchard robot. The navigational accuracy of the algorithm is assessed in a Dutch apple orchard over a distance of 500 m, with the robot driving at two velocities: 0.25 m/s and 0.50 m/s. At both speeds, almost 50% of the observed lateral deviations were lower than 0.05 m from the optimal navigation line. With the use of the PF algorithm, the robot navigated itself between six patterns of tree rows with artificially removed trees. Some lateral deviations exceeded 0.10 m when three adjacent trees were missing in both tree rows. Based on these results, a PF with a laser beam model is an accurate and robust algorithm for the autonomous in-row navigation in semi-structured outdoor environments, such as orchards.
Read full abstract