Research On AGV Autonomous Localization Based On Lidar And Depth Camera

In recent years,with the rapid development and fierce competition of the logistics industry,seeking low-cost,high flexibility AGV navigation system has become the target.In this paper,an AGV system based on low-cost lidar autonomous navigation is proposed.Simultaneous localization and mapping are applied in the paper and using dual-wheel differential mobile robot Turtlebot as experimental platform,which equipped with the domestic laser radar RPLidar and encoder.In order to solve the problem of the encoder data missing and the relocation of robot,the paper give some new ways.The main work is as follows:The frame of SLAM based on filtering is analyzed.In view of the problem that the AGV wheel slippage in the actual scene will cause the odometer data inaccurate,with the robot moving,the pose estimation error accumulated.We applying the Point-to-Line based Iterative Closed Point algorithm to solve this problem.When the error of the odometer exceeds the set threshold,the matching result of two laser frames are used as the recommended distribution,instead of using odometer datas.In the real environment,we did the wheel slip experiments,the proposed algorithm is compared with the original AMCL.The experimental results verify the effectiveness.In addition,in order to solve the problem of robotic kidnapped(or "loss"),this paper proposes a relocation method,which combine the visual features and 2D pose that estimated from laser.During the relocation,that pose is integrated into the current image frame information,and the images collected by the camera are stored in the form of word bag tree to speed up the search and relocation process.When robot need relocation,the current image matches with the image which maintained by the word bag tree,and obtains the current pose considering the similarity and other strategy.In the real-world relocation experiment,our algorithm in structured environment verify effectiveness.Finally,we did some experiment like mapping using the popular laser SLAM,testing the improved algorithm,which in ROS and Turtlebot platform.The results show that positioning accuracy can meet the needs.The accumulated error is 0.362 m in the loop path of 22 m.In structured environment,the accuracy of the relocation process reaches 83.5%,which needs 0.21 s to be recognized,can meet the real-time requirements.