Semantic Map Construction For Cloud Mobile Robots

Cloud robot technology is a combination of cloud computing technology and robot technology,that is,unloading intensive computing tasks on the robot side to the cloud,and using the powerful computing power and rich storage resources of the cloud to reduce the computing and storage pressure of the robot body.Therefore,in the cloud robot system,the terminal equipment can effectively improve the overall performance of the system by calling cloud-related computing and storage services.Mobile robot SLAM,navigation,and obstacle avoidance are typical intensive computing tasks in the field of mobile robots.Traditional SLAM technology is limited to perceiving the geometric structure information of environmental space,and cannot understand the environmental content at the semantic level.Semantic map construction technology can make the cognition of mobile robots transcend the geometric meaning of points,lines,and planes in the environment space,and accurately obtain the semantic information of the environment and its objects.Besides,the navigation and obstacle avoidance strategy based on a deep reinforcement learning algorithm can enable the mobile robot to complete the navigation and obstacle avoidance tasks only according to the location information of the target points in the semantic map after certain "reward and punishment" training.Focusing on semantic map construction and navigation and obstacle avoidance tasks,the main research work carried out in this paper is as follows:Firstly,a semantic map fusion construction method is proposed.Based on this method,a semantic map construction system integrating laser and RGBD camera is designed and implemented.Firstly,a two-dimensional grid map is constructed by using Gmapping algorithm based on filtering and lidar.Secondly,the depth learning algorithm YOLO v3 and RGBD camera are used to identify and locate the target objects in the environment;After that,through coordinate transformation,the target semantic information is mapped to the corresponding position of the grid map and labeled to complete the construction of the semantic map;Finally,experiments in real environment verify the effectiveness of this method.Secondly,a mobile robot navigation and obstacle avoidance strategy based on deep reinforcement learning algorithm PPO is proposed.The input of the network is the original image of the visual sensor,and the output is the moving speed value of the mobile robot.Then,four different simulation scenes world1?world4 are built-in Gazebo,and the complexity of obstacles in the scenes increases in turn;Finally,experiments in World4 show that when the training times reach 3500 times,the strategy can achieve a success rate of more than 75% in navigation and obstacle avoidance tasks.Thirdly,the design,construction,and deployment of the cloud robot system have been completed.Firstly,the overall architecture of the cloud robot system is designed according to the concept of "cloud edge" architecture;Secondly,the cloud robot system is built and deployed,and the system communication mechanism is designed;Finally,through the comprehensive experiment of the cloud robot system in the real environment,the function of the system is tested,and the performance advantages of cloud robot system are verified.