Indoor Lidar-based SLAM Research For Spherical Robot

Autonomous navigation is the key technology of autonomous operation of mobile robot.Currently,the most popular localization technology in autonomous navigation is SLAM(Simultaneous Localization and Mapping).As a new type of mobile robot,spherical robot is good sealing and environmental.Compared with traditional mobile robots,spherical robots are quite different in mechanical structure and kinematic characteristics.General SLAM methods are hardly applied to spherical robots directly,It is difficult and challenging to design SLAM algorithm for spherical robots.This thesis is targeted at the long-term stable operating need of spherical robot in indoor environment.The research is focus on the multi-sensor fusion laser SLAM algorithm for spherical robot platform.The data preprocessing methods are proposed for the sensor data of the spherical robot platform.Based on the error-state Kalman filter,the multi-sensor fusion localization method of the spherical robot is proposed and realized.Based on the occupancy normal distribution(ONDT),a laser SLAM method for spherical robot is proposed.The main research work and contribution of this paper lies in the following:1.The exploration and preprocess of the raw data quality of spherical robot platform sensor.Several experiments are conducted to explore the raw data quality of the sensor.It is found that IMU has large noise and drift.IMU data quality is improved by calibrating the internal parameters of IMU and the robot state is initialized by the IMU data.It is found that the wheel speed odometry of the spherical robot is easily to drift in long time due to the model unfitness and accumulative error.It is found that LiDAR has trailing phenomenon and ground point interference and the quality of LiDAR data is improved by removing trailing outlier and motion distortion.2.A multi-sensor fusion localization method based on error state Kalman filter for spherical robot is proposed.Combined with the mechanical structure and motion model of the spherical robot,the static external parameter calibration and dynamic external parameter calculation of the sensor are realized.The front-end laser odometry based on correlative scan match(CSM)is tested on the spherical robot.To solve the failure of the laser odometry while robot turning,a fusion localization method based on error state Kalman filter is designed,which integrates IMU,spherical robot wheel speed odometer and lidar.Experiments show that this method can get more robust and accurate localization result in spherical robot.3.A laser SLAM method for spherical robot platform is proposed.The theory and framework of graph-based SLAM method is introduced.The construction and update of the map based on ONDT are introduced,and a method of fast filtering dynamic objects in dynamic environment is proposed.The loop detection algorithm based on ONDT and the back-end global pose optimization algorithm are designed and implemented.Experimental results show that the accuracy and robustness of the laser SLAM method designed in this paper meet the requirements of spherical robot.