Research On Visual Navigation Technology Of AMR Robot For Warehousing And Logistics

With the arrival of the intelligent era of logistics and transportation,autonomous mobile robots(AMRs)have greater advantages than traditional AGVs in terms of technology,application and degree of automation,and more and more attention in the field of intelligent logistics.With low cost and abundant information,visual SLAM can greatly improve the adaptability of robots to the environment.Therefore,visual SLAM and route planning technology have become the focus and hot spot in the research of autonomous navigation of logistics robots.Based on the above background,the storage and logistics oriented AMR robot navigation technology is studied.The main research contents are as follows:(1)Aiming at the problem that AMR robot’s positioning accuracy is reduced and the map is prone to "ghost" phenomenon in the complex scene of logistics warehouse with high similarity,high dynamics and man-machine mixing,a visual odimeter optimization algorithm combining lightweight YOLOv5 s and local LK optical flow is proposed.YOLOv5 s network is added to identify potential dynamic objects in the original three-thread of ORB-SLAM2,and dynamic feature points of objects in the anchor frame of target detection are further detected and removed using local LK optical flow method.Only static feature points are retained to complete camera pose estimation.Experimental tests show that in the dynamic scene walking sequence,Compared with the traditional ORB-SLAM2 optimized algorithm,the average error of pose estimation is reduced by 93.26%,the influence of dynamic objects on camera pose estimation is greatly reduced,and the positioning accuracy of AMR robot is improved(2)Aiming at the problem that the traditional ORB-SLAM2 algorithm is difficult to establish sparse map for robot navigation,the optimized camera position and depth image data are converted into point cloud data,point cloud splice is carried out,and dense point cloud map is constructed.The simulation results show that the dense map optimized by visual odometers eliminates the "ghost" phenomenon caused by dynamic objects.The readability and adaptability of maps have been greatly improved.In order to reduce memory occupation,the dense map is stored in octree form.Considering that the warehouse logistics robot moves in a twodimensional plane,only the 3D occupation map of the moving space is mapped into a twodimensional raster map for the subsequent robot navigation.(3)According to the advantages and disadvantages of the global A~* algorithm and the local DWA algorithm,A bidirectional exploration A~* algorithm is proposed to integrate the local DWA algorithm to complete the robot path planning.For the A~* algorithm,a dynamic target point bidirectional exploration strategy is proposed,which not only improves the search efficiency but also avoids the bidirectional disjoint situation.A new heuristic function is constructed to avoid the redundancy of search nodes.Then,the path of A~* algorithm is used as the global guide of DWA algorithm,which not only ensures the optimal global path,but also realizes real-time obstacle avoidance,and effectively improves the autonomous navigation ability of AMR robot.(4)Based on ROS robot operating system and related hardware equipment,an experimental platform for warehousing and logistics AMR robot navigation system was built.The autonomous navigation experiment test was carried out in a real indoor environment with obstacles constraints,and the feasibility and effectiveness of the autonomous navigation system was verified.The robot autonomous navigation system can independently complete the operation of moving and sorting goods,forming a new logistics operation mode of "goods looking for people",effectively liberating the human labor force,and conducive to the formation of a new logistics ecosystem.