Research On Navigation Technology Of Mobile Robot Based On Deep Learning

The problem of navigation and obstacle avoidance of mobile robots is the most difficult research point in the field of robotics.With the advent of the era of artificial intelligence,how to navigate intelligently has become one of the hot research points.The navigation problem is the process by which a mobile robot reaches a target point through its own posture and sensor information and a specific path planning algorithm in an environment with obstacles,and good planning algorithms can effectively improve the feasibility and safety of mobile robots during operation.This paper studies a navigation method that combines Probabilistic Roadmaps(PRM)and deep learning target recognition.It focuses on improving the algorithm's performance in narrow passages and search time,and builds a deep learning model to realize target recognition of obstacles,thus the mobile robot can detect obstacles and avoid obstacles during operation,so as to realize the autonomous and intelligent obstacle avoidance of mobile robots.The main research work of this paper is as follows:(1)This article conducts experiments on a mobile robot platform based on Ackerman's steering constraints,which is equipped with Robot Operating System(ROS),and realizes remote control,path planning and low-level driving under this framework.In this paper,laser SLAM Based on filtering is studied,Aiming at the problem of laser SLAM(Simultaneous Localization And Mapping)self-location prone to errors,this paper uses the method of fusion inertial navigation sensor(IMU)data to enhance the positioning accuracy,thereby increasing the success rate of creating a global map,and then by analyzing the laser radar and IMU data,the data is intercepted to obtain information such as the map,robot attitude and real-time speed required for obstacle avoidance.(2)Aiming at the insufficient adaptability of the PRM algorithm in the case of narrow passages,this paper proposes an improved method based on the optimized sampling strategy to generate sampling points around obstacles and increase the number of sampling points in the narrow passage.Thus we can ensure that there are enough sampling points in the narrow channel in the map,so as to plan a reasonable path,and ultimately improve the adaptability of the PRM algorithm in a complex environment.(3)In order to ensure that the mobile robot recognizes obstacles in real time,considering the limited computational cost,this paper builds and improves the YOLOv3 deep learning model,tailors the model,and lightweight deep learning model to improve the recognition speed while ensuring that the accuracy will not be greatly reduced.Build Darknet_ROS in the mobile robot platform,and combine the improved YOLOv3 model with ROS,so as to quickly and efficiently realize the combination of deep learning and mobile robot software and hardware.Finally,by using the image information collected by the camera and identifying specific obstacles through the deep learning model,real-time local obstacle avoidance is realized by combining the effective information obtained by the TEB(Timed-Elastic-Band)algorithm and the robot sensor according to the position of the obstacle.In summary,this article studies the navigation technology based on deep learning,improves the traditional laser radar mapping method,proposes an improved method for the defects of the PRM algorithm,and builds a deep learning model and formulates obstacle avoidance strategies.Finally,a navigation obstacle avoidance test was carried out on the mobile robot platform,so that the mobile robot can accurately identify obstacles and avoid obstacles.