Research On Cloud Robotics Navigation Based On Visual SLAM

Visual SLAM is becoming a hot spot mobile robot navigation.When excuting SLAM,navigation tasks,traditional robots have many shortcomings,Huge data processing and computing tasks,which consumed much storage and computing resource.There is a high demand for storage and computing resources,which is in contradiction with the limited resources of the airborne hardware.Based on the advantages of cloud computing's computing power and resource sharing mechanism,this thesis combines robot with cloud computing to research the mobile robot's navigation using visual SLAM.This thesis has strong novelty and high scientific value.In the indoor unknown environment,this thesis utilized the ORB SLAM algorithm to realize the 3D cloud point map reconstruction.The steps consists of sensor reading,visual odometry,back-end optimization,loop closure and graph optimization.What's more,the map precision is 0.01 m.After the indoor environment point cloud map is generated through the ORB SLAM,then converted it to octomap for navigation.In the design of global path planning algorithm,based on the Rapidly-exploring Random Tree(RRT)path planning algorithm,this thesis introduced the path cost function heuristic A*algorithm to improve the random node generating and step function.So the algorithm can search the destination more effective and stable.Simulation has shown the performance of the improved path planning algorithm.By means of ORB-SLAM's keyframe and bag of visual words technology,the system solves the relocalization during the autonomous navigation.In the design of the cloud robotic navigation system,the system is divided into three sides,the robot side,the cloud side and the user-end side.The robot side consists of Kinect,turtlebot2 base and low cost embedded board computer.The cloud side runs in the Priviate cloud server with a ROS-based linux virtual machine.What's more,a django web server is deployed in the cloud.For the user-end side,the system has a web application as the virtualization layer for displaying the messages for remote human-robot interaction.The messages include the robot's real-time kinematic trajectory,velocity,position and the video stream.The cloud side and the user-end side of the system are implemented with a B/S structure.According to the concept of the cloud computing sharing mechanism,the robot resources are provided to users in the form of web services,which provide a more economical and friendly human-computer interaction.Experiments have realized the functions of remote manipulation,mapping and autonomous navigation and the results have shown the real-time performance and feasibility of this systemFinally,this thesis summarizes the research work,and prospects the future research contents and directions.