基于激光雷达导航的玉米喷药机器人设计与试验

Design and Experiment of Corn-spraying Robot with LiDAR Navigation

摘要:

针对现有玉米喷药机器人存在转向响应慢、作物行检测方法以及跟踪控制器稳定性差等问题，设计一种基于激光雷达导航系统的四轮驱动、差速转向式线控底盘喷药机器人。首先设计机器人整机结构，根据工作原理对关键部件进行设计。提出一种基于3D激光雷达的作物行检测方法，经过点云预处理和地面点云滤波得到机器人前方的作物点云，根据点云在横向坐标轴的分布数量划分不同的作物行，并利用分段几何中心拟合作物行中心线。同时，设计一个双输入单输出的模糊控制器，以作物行中心线获取的偏航角和横向偏差为输入，利用49条模糊规则和Mamdani法进行模糊推理，再通过重心法将输出量解模糊为线控底盘两侧车轮轮速差。在苗期玉米田进行机器人行驶性能试验和导航性能试验。试验结果表明，机器人能够爬越坡度20°以下斜坡，差速原地转向360°时几何中心位置平均偏差为7.66 cm，具有足够的驱动力和较好的转向灵活性;激光雷达检测三叶期和小喇叭口期的玉米作物行时，平均误差角分别为0.93°和0.85°，平均检测时间为0.031 s，据此确定的定位信息具有较高的精度且满足实时性;通过定位信息和模糊控制器跟踪作物行时，机器人平均跟踪误差为0.061 m，标准差为0.038 m，能够满足苗期玉米田的自动导航需求。

Abstract:

Aiming at the problems of slow steering response, crop row detection methods, and poor stability of tracking controllers in existing sprayers, a four-wheel-drive, differential-steering spraying robot was designed based on a light detection and ranging （LiDAR） navigation system. The whole structure of the robot was firstly designed, and the key components were designed according to the working principle. Then a crop row detection method based on 3D LiDAR was proposed. This method involved obtaining the crop point cloud in front of the robot through point cloud preprocessing and ground point cloud filtering. Subsequently, different crop rows were identified by analyzing the distribution of point cloud data across the transverse coordinate axes. The centerlines of the crop rows were then determined by fitting segmented geometric centers. Meanwhile, a dual-input single-output fuzzy controller was designed to use the yaw angle and lateral deviation obtained from the centerlines of the crop rows as inputs. The controller performed fuzzy inference by using 49 fuzzy rules and the Mamdani method. The outputs were then defuzzified into the differential wheel speeds for the wheels on both sides of the wire-controlled chassis by using the center-of-gravity method. Finally, the robot driving performance test and navigation performance test were conducted in the seeding cornfield. The results showed that the robot can successfully climb slopes over 20°, and the average deviation of the geometric center was 7.66 cm when performing a turn at differential speed. This indicated that the robot possessed adequate driving force and excellent steering flexibility. When LiDAR detected corn crop rows at the three-leaf stage and the small trumpet stage, the average error angles were 0.93° and 0.85°, respectively, with an average running time of 0.031 s. Utilizing this localization information, the robot achieved an average tracking error of 0.061 m with a standard deviation of 0.038 m when navigating the crop rows through the fuzzy control algorithm. This level of accuracy can meet the requirements for automatic navigation in corn fields during the seedling stage.