Research On Navigation System Of Indoor Service Robot Based On Multi-sensor Fusion

With the strong support of the country in the field of artificial intelligence and the active participation of scientific research personnel,the development and application of indoor service robot has a very broad prospect,and there have been very considerable achievements.The Simultaneous Localization and Mapping,known as SLAM,of service robots in unknown environments,is a significant problem in the robot field and a prerequisite for mobile robots to complete instructions.Currently there are two according to the sensors: LIDAR SLAM and Vision SALM.LIDAR can perceive the obstacle in the located plane and build the plane map,but it lacks detailed 3d information.Vision sensors are divided into monocular camera,stereo camera and depth camera,which can perceive 3d scene but are affected by external environment such as light intensity.It is difficult for a single sensor to satisfy the robot's scene adaptability and various applications,so the multi-sensor fusion in this paper can enhance the robot's scene use ability and make its system more robust.In this paper,2D LIDAR and RGB-D camera were used as the experimental platform,and the information of the sensor was integrated based on ROS to obtain more accurate two-dimensional raster map,which was used in the path planning.The research work of this paper is mainly divided into the following parts :(1)The differential motion model of wheeled robot and the perception model of 2D lidar and depth camera are established.(2)Description of laser and visual SLAM technology based on pose graph optimization: visual SLAM uses sparse feature method to extract ORB features for feature matching and use EPnP to obtain motion estimation.with the laser SLAM,use the two-stage correlation matching with Coarse to fine to make point set registration.The back end uses the pose diagram to carry on the optimization.(3)A scheme of raster map fusion based on Bayesian estimation is proposed.(4)use the improved A* algorithm for global path planning,and use the dynamic window method to complete local obstacle avoidance.(5)Design experiments are carried out to analyze and verify the effectiveness of the proposed method.