| Autonomous navigation of unmanned forklifts in warehouse is an important technology required for logistics distribution systems to realize automation and intelligence,where real-time localization and path planning are key components.As for localization,visual SLAM and tag localization are current popular technologies.Traditional visual SLAM does not rely on artificial environment with low hardware cost,but the inadequate accuracy and the serious accumulated error are endured in extensive warehouse.The artificial visual tag enables localization with high accuracy,but the layout is limited in warehouse and the observation is difficult for the unmanned forklifts.As for path planning,on the special condition in warehouse environment,the algorithm needs to run efficiently,reduce path cost and generate smooth trajectories,which are not well met in existing algorithms.To address the above issues,the paper researches real-time visual localization and global path planning on unmanned forklifts in the logistics warehouse.The main contents are as follows:1.Design of a visual recognition and localization method for tags in warehouse shelf environment.For the difficulty in tag recognition in warehouse shelf environment,a 2-layer filtering method based on heuristics is proposed.AprilTag is selected as the visual tag.Firstly,images go through operations including threshold segmentation and boundary extraction,and tag regions are preliminarily filtered with a loose matching strategy.Then,combined with the unique depth information from the system,false positive tags are eliminated and the further filtering is completed.Next,based on the pinhole camera model,the coordinates of the identified tag corner points are used to solve the camera pose with the direct linear transformation method,achieving the tag localization.The field experiments show that the 2-layer filtering method is able to effectively identify the tags on the shelves and achieve the high-accuracy centimeter-level visual localization for unmanned forklifts.2.Design of a visual localization algorithm for unmanned forklifts in the logistics warehouse.The TGF-SLAM(Tag-guided Feature-point SLAM)algorithm performing visual localization by both AprilTag and feature points is designed,where the problem of the accumulated error in feature point localization is tackled using the absolute position information provided by tags.Firstly,the scheme is analyzed,where the depth camera is selected as the sensor and the idea of "tag first,map refined" is established.Then the general framework of the algorithm is built and the graph optimization methods are designed for tags and feature points,including single frame optimization,small-area map optimization,long-section pose optimization among tags and loop pose optimization.Next,the specific operation mechanism of tag-guided localization is designed to harmonize the localization of tags and feature points.The field test proves that our TGF-SLAM is able to basically eliminate accumulated errors and the decimeter-level localization accuracy meets the system requirement.3.Design of a global path planning algorithm for unmanned forklifts in the logistics warehouse.The path planning is divided into two phases,including search and optimization.In the first phase,the RF-RRT(R-tree-based Fast RRT)algorithm for path search is designed,where R-tree is utilized to spatially cluster map points to accelerate the search of near points,complemented with strategies of bidirectional search and greedy growth to improve the efficiency.In the second phase,the path optimization algorithm is designed with segmented third-order Bezier curve as the smoothing model and the simulated annealing process as the framework,including cooling layer,path adjustment layer and basic calculation layer from top to bottom.In addition,to ensure the safety of path planning,morphological expansion is used to transform the warehouse map into C-space considering the obstacle avoidance distance.Simulation experiments demonstrate that our algorithm is able to achieve fast path search,generate cost-effective and smooth trajectories for unmanned forklifts in the warehouse map. |
