Research On Indoor Scene Understanding And Navigation System Of Wheeled Service Robot

With the intensification of population aging and the lack of service workers,service robots are developing rapidly.However,there are currently some bottlenecks in the usage of service robots.One aspect is that the service robot has insufficient scene perception and understanding in a complex environment.Because it cannot obtain the semantic information of the environment,it has no ability to complete higher-level tasks.The other aspect is that,when there are more obstacles in the environment,poor light conditions,or requiring high precision,the laser sensor and the depth camera sensor both have some limitations in fulfilling the requirements of navigation in complex structured dynamic environment.In order to solve the above problems,this article develops a service robot indoor scene understanding and navigation system,and researches the application of visual target detection technology,simultaneous positioning and mapping technology,and path planning technology on mobile robots.The main contents are as follows:(1)In order to improve the intelligent level of service robots,this article uses deep learning technology to develop a robot scene perception system.By using YOLOv3 model,the understanding and detection of scene objects are realized.It meets the requirements of real-time and accuracy of system understanding.After obtaining the target area in the image,the distance information of each target is obtained based on depth camera.Finally,the target detection and ranging are packaged as different ROS nodes for deployment on an industrial computer.The integration with the robot navigation,positioning,and grasping modules is realized,which provides conditions for the realization of advanced tasks(2)This paper studies laser SLAM and visual SLAM technologies.On this basis,a multi sensor fusion mapping scheme based on Bayesian estimation is proposed.The fusion algorithm can obtain more detailed scene information than a single laser mapping.It makes up for the limitation that the single-line laser can not recognize the obstacles outside the plane.The fusion algorithm obtains higher accuracy than visual sensor,and provides more detailed environmental information for path planning.(3)On the basis of the environment grid map,the improved A*algorithm is used for global path planning.The dynamic window method is used to realize the robot avoiding obstacles.On the basis of the above research,the motion model of omnidirectional mobile robot is studied.The mobile grasping robot experimental platform is built to verify the scene understanding scheme,the fusion mapping scheme,and the path planning scheme.