Research On Laser/Vision-Guided Compound Mobile Robot

The compound mobile robot is a kind of multifunctional robot that combines an articulated robot and a wheeled mobile robot platform.It overcomes the limitation of the working space of the robotic arm through the free movement of the mobile platform and is widely used in logistics handling,intelligent services,industrial production and field exploration,etc.Aiming at the positioning,navigation and target grasping of compound mobile robots,this paper studies the software and hardware architecture of control system,kinematics modeling and simulation,laser/vision fusion navigation and positioning,path planning as well as technology and realization of object grasping based on vision and deep learning and their implementation.The main research contents include:Based on the analysis of the functional requirements of the compound mobile robot,the mechanical structure design and hardware selection are carried out,the software and hardware scheme of the control system is planned,and the control system architecture based on the hierarchical finite state machine is proposed.The hardware of the mobile platform adopts the distributed control system architecture of the GPU server and the vehicle-mounted controller.The GPU server realizes the data acquisition and processing of sensors such as dual-channel lidar,odometer and stereo vision camera;the vehicle-mounted controller is based on the STM32 embedded control system to realize the whole motion control to mobile platform;software platform adopts UBUNTU and ROS operating system.To meet the reliable and robust control requirements of articulated manipulators,wheeled mobile platforms and multimodal sensing complex heterogeneous systems,a hierarchical finite state machine with a main state-sub-state structure is constructed,and state transitions are driven by specific events.Aiming at the high-precision positioning requirements of environment cognition and target manipulation of compound mobile robots,three-dimensional dense reconstruction is realized based on real-time appearance mapping,a laser/visual composite positioning method and a positioning accuracy evaluation method based on circular probability error are proposed.In appearance-based real-time mapping,loop closure detection adopts a memory management mechanism that selects landmark points according to the access frequency,and then realizes point cloud splicing.The laser/vision hybrid localization firstly adopts rough localization based on lidar guidance and a posterior probability model.Then Apriltag signpost was used to compensate visual localization.In order to evaluate the positioning accuracy,the ellipse fitting of the target position is used to get the circular error probable.The experimental results show that the method can improve the positioning accuracy significantly.According to the target manipulation requirements of the compound mobile robot,t the kinematics modeling of the joint robot is carried out based on the D-H method.To achieve integrated integration,the description and specification models of the joint robot and the mobile platform are created respectively.And the forward and inverse kinematics solutions and simulation verification are also carried out.The prototype of the composite mobile robot was developed and systematically tested focusing on path planning and target grasping.The global-local path planning is carried out by using recursive redundant node elimination and sampling backward exception processing mechanism;and target detection is realized based on SSD neural network,and the target centroid coordinate transformation is calculated to determine the grasping pose.The related algorithms are verified by robot grasping experiments.