Research On Autonomous Navigation Of Omnidirectional Mode Mobile Robot For Environmental Detection

With the further development of science and technology,as well as the ever-changing demand of robots for society,the requirements for robotics are correspondingly improved,and more research is needed on the autonomous navigation technology of robots.This thesis studies the technology of autonomous navigation of omnidirectional mode mobile robots for the autonomous navigation needs during environmental detection.For the dangerous environment and multi-obstacle scenes that need to be detected,based on the laboratory self-developed multi-sports mobile robot,the kinematics model in omnidirectional mode is built and its autonomous navigation system is designed.The autonomous navigation system which based on ROS is divided into three layers.They are the upper decision layer,the intermediate communication layer and the lower control layer,the hardwares that required for the three layers are separately selected and designed.For the task of autonomous navigation,traditional navigation method and navigation method based on deep reinforcement learning are proposed.With the traditional navigation method,for the requirements of map construction,it is necessary to analyze the characteristics of Hector's algorithm and the characteristics of Gmapping algorithm and select the algorithm with the best performance.Aiming at the motion characteristics of omnidirectional movement,the simulation results of the global path planning algorithm A* for implementing autonomous navigation and the dynamic window method for introducing lateral velocity sampling of the local path planning algorithm are respectively carried out.The traditional navigation relies on a priori map,so a navigation method based on deep reinforcement learning is proposed.On the basis of understanding the characteristics of deep learning and reinforcement learning,the research on deep reinforcement learning is introduced and try to apply Deep Q-Network to the mobile robot navigation.The various elements of the algorithm are designed.In order to verify the effectiveness of the algorithm,the simulation training scenario of path planning is designed and simulated.Based on the above work,the traditional navigation method was simulated and the deep reinforcement learning training tool was designed with the simulation scene and mobile robot under Gazebo.The tool is used for deep reinforcement learning training and simulation.After the verification algorithm,two autonomous navigation methods are migrated to the physical robot.Experiments show that under the traditional navigation method,mobile robots can have better autonomous navigation performance while playing a better dexterous motion capability.The navigation method based on deep reinforcement learning is applied to multi-motion mode mobile robots in omnidirectional mode,and autonomous navigation can be achieved without a priori map.