Research On Environmental Perception And Motion Planning Of Self-propelled Plant Protection Robot

In view of the low efficiency,low intelligence,poor safety and reliability,and high labor intensity of the plant protection machinery in the plant protection process,the environmental perception,path planning and motion control technology are used to study and design the plant protection robot navigation system.The lidar is used to sense the environment vineyard.According to the different characteristics of the sensor feedback data in different coordinate systems,two schemes of filtering,classification and fitting are designed to fit the ridge line and navigation line.The relative pose relationship between the robot and the navigation line can be determined.The robotic kinematics model and the navigation line tracking model are structured to obtain the desired rotation angle of the motion control,and the underlying motion controller is used to realize the autonomous motion of the plant protection robot.This paper discusses the research foundation of the project,analyzes the development status of agricultural robot in China and abroad,and combines the actual working environment of the vineyards in China.According to the environment perception,control decision as well as motion execution,the overall research plan of the system was designed and the technical route was formulated.The autonomous plant protection robot system can be divided into three modules in the overall structure.The first part is the environment sensing layer,which mainly detects the surrounding environment and performs self-positioning.The second part is the decision layer.The main work is to determine the target motion trajectory and obtain the desired action;the third part is the motion execution layer,according to the target corner,construct a closed-loop motion controller to achieve accurate motion following.Single-line lidar is used to detect the vineyard environment where the plant protection robot is located.As the eye of the robot,the lidar can determine the distribution of the ridges and capture the local environmental information.The point cloud data scanned by the lidar can effectively reflect the arrangement information of the ridges and branches of the vineyards,but the noise points will cause a large interference to accurate judging the position of the ridge line,and the pre-processing of the point cloud data becomes an indispensable link.According to the smooth single-peak distribution of the lidar point cloud data in the polar coordinate system.Based on the original data,an envelope model based on the moving average filter band is constructed to complete the filtering and denoising.Combined with the peak points in the point cloud data,the classification criteria of the ridge lines on both sides are determined.The minimum variance method is used to fit the classified point cloud data,obtain the exact position of the ridge line,extract the center navigation line,and determine the pose relationship between the lidar and the center navigation line.Aiming at the significant trend of the point cloud distribution in the Cartesian coordinate system,which shows effective points usually attach to ridge line,the threshold and Kalman filter is designed to complete the filtering and denoising of the original data.The SVM mechanism is used to realize the data classification and the model of multi-support with the matching weight to determine the position of ridges is constructed.Using RASANC based on the angle division principle to fit the ridge line and extract the center navigation line has good effect.The kinematics model of the plant protection robot is constructed,and on this basis,the target navigation line tracking model of the robot is designed to determine the next desired action.According to the pose relationship between the robot and the central navigation line outputted by the sensory layer,combined with the actual mechanical structure and motion system of the plant protection robot,the tracking model of the robot to the target navigation line is constructed.The motion trajectory planning of the plant protection robot is also completed.According to the desired action of the plant protection robot relative to the central navigation line,the steering controller of the plant protection robot is designed to realize the accurate follow-up of the desired action.The execution layer,motion controller are designed based on the fuzzy PID to complete the accurate follow of the desired corner.The filtering algorithm,classification method and fitting strategy of the sensing layer are verified by experiments.The kinematics model of the plant protection robot,the navigation line tracking model and the steering controller in the execution layer are tested.The experimental results show that on the one hand,the plan of Multi-support vector based on support vector machine for discriminating safe position of ridge and extracting navigation line has the average error of the offset which is 4.2mm,the average error of the yaw angle which is 0.69°,and the average consuming time which is 2.06044 s,which can not meet the requrirement of engineering application;on the other hand,the envelope model based on the moving average filter band can effectively filter out the noise.Iterative search peak classification method has the average success rate of effective classification is 99.7%.The minimum variance method based on the angle division principle is used to extract the navigation line.The average error of the offset is 25.1mm,the average error of the yaw angle is 1.18°.The designed navigation line tracking model and the steering controller in the vineyard environment have an average time constant for reaching steady state is 8.21 s,and this plan has the average consuming time of 1.5ms,which can meet the practical application requirements of the autonomous operation of the robot in the vineyard environment.