Orchard robot navigation in unstructured environment based on density adaptive RANSAC

An orchard navigation scheme based on multi-sensor fusion was proposed to solve the problems of weak signal and poor positioning effect caused by tree occlusion in the GPS navigation process of orchard robot .

High-precision 3D point cloud data were collected by 16-line lidar, point cloud was preprocessed by Voxel grid filter algorithm, point cloud density was reduced and discrete points were removed, fruit tree rows were clustered by Euclidian algorithm, and the straight lines of fruit tree rows were fitted by improved random sampling consistency (RANSAC) algorithm. According to the relationship of parallel lines, the navigation line was calculated and integrated with inertial measurement unit (IMU) for high-precision positioning of orchard robot. Based on differential steering and pure tracking model, the goal of autonomous navigation and automatic line wrapping of orchard robot was realized.

After the data fusion of lidar and IMU, the accurate position and pose of the robot were obtained. Compared with the deviation produced by the least square method and the traditional RANSAC method, the lateral deviation based on density adaptive RANSAC method was less than 0.1 m and the heading angle deviation was less than 1.5° when the robot was operating in the orchard at the speed of 0.8 m/s. The deviations were the minimum in the three methods. However, when the robot speed increased to 1.0 m/s, all the deviations increased obviously.

The orchard robot navigation technology based on multi-sensor fusion proposed in this paper is suitable for most standardized orchards and has important promotion value.