Research And Application Of Key Technologies For Mobile Robot Intelligent Grasping Using RGB-D Sensor

With the development of artificial intelligence and industrial robot technology,intelligent mobile robots have become an important means of advanced intelligent manufacturing and services.Grasping is a key operation step in mobile tasks such as handling,sorting,and assembly.Therefore,it is of great significance to research efficient,accurate,and highly reusable grasping methods for the intelligentization of industrial robots.Aiming at the autonomous grasping operation of mobile robot in warehousing and logistics environment,the key technologies of mobile grasping based on the combination of vision and laser are researched,and the autonomous grasping application software for mobile robot is developed in this dissertation.Firstly,in view of the robustness and poor reusability of traditional object detection algorithms,considering the characteristics of the structured environment and the real-time requirements of grasping tasks,an improved object detection algorithm based on YOLOv3-Tiny is proposed,which realizes real-time detection of the object to be grasped.Aiming at the low detection accuracy for small objects using YOLOv3-Tiny algorithm,several measures such as increasing the detection scale,introducing the space pyramid pooling module and optimizing the anchor acquisition method are proposed,significantly improving the detection accuracy for small objects and increasing the detection speed.Secondly,aiming at the structural characteristics of the two-finger parallel gripper and the grasping requirements,a plane grasping pose estimation algorithm based on the combination of depth information and image moments is proposed.The modeling and calibration of the vision system are presented and realized,including camera modeling,camera calibration,handeye calibration and image registration.Considering the presence of random noise and partial loss in depth images,a cascaded image processing framework for image restoration and enhancement is designed to obtain the depth information of the captured points.The grasping angle of the target object is calculated by using the image invariant moment theory.Finally,the depth information and the grasping angle are merged to obtain the plane grasping pose description in the camera coordinate system,which is mapped to the base coordinate of the manipulator.Finally,a hierarchical and coordinated planning scheme for mobile robot is proposed.For the mobile platform,a coarse-then-fine planning strategy is proposed.Base on the environment map,the mobile platform navigates globally to the capture point using classic A* algorithms,and then optimizes the local pose based on the operability.For the robotic arm,an improved planning algorithm based on RRT-CONNECT is proposed.By adding intermediate nodes,introducing corner constraints and improving the sampling step size,the planning success rate is improved.Based on the above-mentioned key technology research results,a mobile capture system based on the fusion of RGB-D camera and lidar is designed and constructed.What's more,a mobile capture application software is developed.Finally,the physical experiment of grasping the target object verifies the effectiveness of the algorithm proposed in this paper.