Toward Autonomous Navigation for Agriculture Robots in Orchard Farming

The agricultural sector is increasingly embracing automation for key operations such as planting, harvesting, and pesticide spraying. However, the effectiveness of these systems often depends on specific environmental conditions. Navigating through orchards with automated systems using RTK-GNSS or row tracking can be challenging due to factors such as uneven crop distribution, environmental obstacles, and signal instability under tree canopies. In this study, we present the design and development of an autonomous agricultural robot intended to navigate orchards and assist with pesticide spraying tasks. We propose an innovative navigation framework that utilizes visual information from two stereo cameras to identify crops. The localization system integrates sensor fusion, combining RTK-GNSS, visual odometry, and IMU data for enhanced accuracy. Additionally, we introduce a novel hybrid controller based on the Dynamic Window Approach (DWA), enabling the robot to efficiently follow fruit tree rows even when positioning signals are disrupted or trees are not detectable in certain areas. This eliminates the need to switch between control modes such as row following or reference trajectory tracking. The proposed method has been implemented on ROS2 and validated through simulations in various complex environments. Furthermore, real-world experiments were conducted in actual fruit orchards, demonstrating the efficacy of the proposed system. The hardware and software modules are also detailed in this study.