Research On Unstructured Environment Navigation Technology Of Agricultural Robot Based On Multi-sensor Fusion

For ins and outs of agricultural machinery based on unstructured environment and orchard independent operation,put forward a multiple sensor fusion laser SLAM navigation scheme,RANSAC navigation scheme based on adaptive density,and discusses the navigation of agricultural robot control strategy,which laid a foundation for realization of ecological farm unmanned,concrete research content is as follows:(1)For the data fusion of multiple sensors,this paper chooses the combination of Li DAR and IMU.In order to improve the precision of data,it is necessary to calibrate the two sensors.First,establish their own coordinate system,including world coordinate system,Li DAR coordinate system and IMU inertial navigation coordinate system.Secondly,the transformation matrix of the coordinate system is mainly composed of translation matrix and rotation matrix.Then,the sensor time synchronization,align the timestamp,unified description of an object in the coordinate system position;Finally,the data is fused with the internal parameter errors of sensors.(2)Aiming at the problem that agricultural robots can’t get in and out of storage under the traditional GPS navigation system,a laser SLAM navigation scheme based on the fusion of Li DAR and IMU is proposed.In an unknown environment,Cartographer algorithm based on graph optimization was used to acquire Li DAR information,IMU information and odometer data to construct feature raster map with resolution of 5cm.Two points were optimized for the traditional Cartographer algorithm.One was to improve the quality of input point cloud through multiple filtering algorithm.Second,multi-source fusion pose estimation algorithm is used to reduce the error.The error of Cartographer algorithm constructed by the improved Cartographer algorithm is less than 10 cm,which significantly improves the quality of SLAM mapping(3)For the autonomous operation of agricultural robots in orchards,this paper proposes a density adaptive RANSAC(random sampling consistency)navigation system.The 16-line Li DAR was used to collect high-precision THREE-DIMENSIONAL point cloud data,and the Voxel Grid Filter algorithm was used to preprocess the point cloud,reduce the density of point cloud and remove discrete points.The fruit tree rows were clustered by Euclidian algorithm,and the improved RANSAC algorithm was used to fit the straight lines of fruit tree rows.According to the relationship of parallel lines,Calculate the navigation line;And IMU is used for high precision positioning of agricultural robots.Based on differential steering and pure tracking model,the goal of autonomous navigation and automatic line wrapping of agricultural robot in fruit trees is realized.After the data fusion of Li DAR and IMU,the accurate pose of the robot was obtained.When the robot was working in the orchard at a speed of 0.8m/s,the lateral deviation generated by the density adaptive RANSAC method was less than 0.1m and the heading Angle deviation was less than 1.5°,compared with the deviation generated by the least square method and the traditional RANSAC method.Are the minimum values of the three methods.(4)For the navigation control strategy of agricultural robot,pure tracking algorithm is selected in this paper for the navigation in the orchard.Firstly,the basic motion model of the robot is analyzed.Secondly,the basic principle of pure tracking algorithm is described.Then,the trajectory of the agricultural robot is simulated to determine the appropriate forward-looking distance and complete the parameter optimization.Finally,the stability control of agricultural robot is realized by combining differential motion model and pure tracking algorithm.In this paper,the navigation control framework of LASER SLAM is selected to accurately locate the position of agricultural robot by combining the data of Li DAR,IMU and encoder,and A* algorithm is used to plan the appropriate global path to realize the navigation of agricultural robot in and out of the agricultural machine depot.In addition,the dynamic window method is used to optimize the local route so as to avoid the moving obstacles.According to the analysis of several experimental results,the navigation and positioning error of agricultural robot is less than 8cm when entering and leaving the warehouse,and the obstacle avoidance effect is very good within the visual range.So,this one difficult problem about agricultural machinery out of storage has been effectively solved.