Positioning and map construction for agricultural robots integrating odometer information

Abstract: A relatively low-cost and low frame rate lidar can be very popular to serve as the main sensor of Simultaneous Localization and Mapping (SLAM) in the mainstream agricultural robots at present. The lidar can scan the environment, while the pose of the robot can also change in the SLAM process. However, motion distortion and mismatching can often occur for the environment mapping, because one frame of lidar data can be obtained under the same pose of the robot by default. The resulting errors in the SLAM map can directly determine the accuracy of the automatic navigation of the robot. In this study, a commonly-used classical SLAM Gmapping was utilized to integrate the odometer information, to reduce the motion distortion of lidar with a low frame rate. The displacement and angle of the robot were directly measured with high accuracy of local position using an odometer, one type of important pose sensor in robots. An approximate odometer pose was also matched for each laser point in a frame of data that was scanned by lidar, according to the odometer information of high frequency. Among them, the odometer information was considered to be the collected laser points, thereby obtaining their coordinates after removing the motion distortion. Finally, the data was re-encapsulated to reduce the motion distortion of low frame rate lidar data on the map construction in this frame. A SLAM mapping test was also carried out to verify the improved Gmapping in a maize field and banana garden using a robot equipped with RPLIDAR A1 lidar with a scanning frequency of 5 Hz. The experimental results showed that the average absolute error of the improved Gmapping was 0.01 m in the maize field with a length of about 12 m, 0.05 m lower than that of the original one (0.06 m). In the banana garden area with a length of 24.43 m, the average absolute error of the improved Gmapping was 0.07 m, 0.39 m lower than that of the original one (0.46 m). Consequently, the mapping accuracy of the improved Gmapping was higher than before in the agricultural environment, indicating that the motion distortion of lidar with a low frame rate can be removed effectively.