Autonomous Navigation and Obstacle Avoidance in an Orchard Using Machine Vision Techniques for a Robotic Mower

Abstract

This study proposes an autonomous navigation system for a robotic mower using two RGB-D cameras and an inertial measurement unit (IMU) without global navigation satellite systems (GNSS) in an orchard. In this research, a convolutional neural network (CNN) was used to recognize obstacles, and visual simultaneous localization and mapping (VSLAM) was used to estimate the mower’s position. In addition, landmarks were placed in the environment to compensate for the self-position estimation error. An autonomous navigation experiment in a simulated orchard resulted in the difference between the estimated position and the true position of 0.30 m, and the robot was able to drive in the area without colliding with obstacles due to obstacle recognition by the RGB-D cameras.