Multi-station Preparation-oriented Mobile Robot Navigation And Target Pose Estimation

The robot has the advantages of high efficiency,low consumption and flexibility,and is widely used in the field of warehouse storage materials.However,the storage environment is limited by the large space,the wide variety of goods and the disorderly placement of goods.Robots are difficult to sort and carry goods.At present,most of the semi-automatic equipment used in the field of warehouse storage materials still requires manual cooperation.The mobile robot arm takes into account both flexibility and maneuverability,combined with SLAM technology and visual positioning technology,and can be well used in the field of warehouse storage materials.Aiming at the problems of insufficient real-time performance and low positioning accuracy of the mobile robot arm's pose estimation system during multistation material preparation,this paper has developed a set of intelligent mobile robot arms oriented to multi-station material preparation by fusing data from multiple sensors.The tedious and important task of material preparation at work stations is as follows:(1)Based on the LineMod format data set,this paper proposes to use the smallest outsourcing cube of the target object to replace the pose of the target object.By constructing the coordinate conversion relationship,the two-dimensional information corresponding to the three-dimensional information is solved by the mapping function,and the information is used as the convolutional nerve.The data input of the network trains the deep learning model,thereby avoiding the production of three-dimensional models and improving the efficiency of data set production.(2)Aiming at the problem of low real-time performance of the mobile robot pose estimation system,after in-depth research on the characteristics of the target,the YOLO6 D algorithm is optimized by changing the convolutional neural network structure.The experimental results show that the improved network recognition accuracy is as excellent as the original network,3.11% higher than the accuracy of BB8,but the running speed is nearly12 times that of the BB8 algorithm and more than 17 times that of the Brachmann algorithm.In NVIDIA Ge Force RTX 2060 The running speed on the GPU can reach 35 FPS,which is suitable for real-time processing.(3)Aiming at the problem that the initial cost map is not automatically updated when the mobile robot is navigating,a method of automatically cleaning the cost map is proposed.Load the cost map cleaning instructions into the navigation startup file,and automatically clean the cost map every time the navigation program is started,thereby improving the success rate of fixed-point navigation.In the experiment,the cost map and the cost map are cleaned up when the map is initialized before navigation.Does not clean up the cost map.The experimental results show that the success rate of fixed-point navigation for cleaning the cost map is 46.7percentage points higher than that of not cleaning the cost map.(4)This article uses A~* global path planning algorithm and DWA local path planning algorithm.Based on the ROS robot operating system,it integrates lidar,odometer,and IMU data to achieve the goal of automatic material preparation for mobile robotic arms in a multistation environment.Experiments show that the maximum positioning error of the fixed-point navigation of the mobile robotic arm is 4mm,which meets the requirements of use.(5)Based on the tasks to be completed by the mobile manipulator,an experimental platform was built,and a corresponding software system was designed and developed.The sports camera model was established,and the hand(manipulator)eye(camera)relationship calibration was completed.A complete verification experiment was designed.The content of the experiment included fixed-point navigation,identification,gripping,obstacle avoidance,and target placement of mobile robots.Experimental results show that the average accuracy of the robot's fixed-point grasping reaches 90.43%,and the average success rate of performing continuous motion grasping reaches 80%.Therefore,the research in this paper provides a feasible solution for the navigation and pose estimation of the mobile manipulator,and lays the foundation for the final application.