Research On 2D SLAM Method In Dynamic Scene

Simultaneous localization and mapping(SLAM)is one of the key technologies for mobile robot autonomous navigation and enviroment exploration.There are many excellent works on SLAM research.However,most of the researches focus on the static and rich texture environment.How to improve the SLAM performance in the dynamic,weak texture scene is still a challenging task.Based on the existing 2D laser SLAM methods,considering with the characteristics of dynamic and weak texture environment,this paper proposes the corresponding improvement measures in map building,data association,closed loop detection,back-end optimization,improved the performance of SLAM in dynamic and weak texture enviroment.The work of this thesis mainly includes:1)Multi-sensor fusion based data association method is addressed.With the framework of the Extended Kalman Filter(EKF),we combine the 6DOF pose estimation with the 3DOF Lidar scan matching results,and take the fused results as the initial value of the second scan matching to achieve more accurate initial pose in the dynamic and weak texture scene.The submap-based scan matching method is used to reduce the error accumulation and enhance the ability to resist the interference from moving target,and the Levenberg-Marquardt(LM)method is adopted to solve the nonlinear optimization problem of scan matching to obtain better robustness.A 2D grid map is used to represent the workspace.The map resolution and updating value are set empirically by experiments.The Bicubic interpolation method is used to make the grid map continuous and to improve the accuracy of scan matching.2)Sparse Pose Adjustment(SPA)is exploited to the back-end optimization.All submap and the poses of Lidar are optimized by closed-loop detection constraints.In order to reduce the influence caused by wrong closed-loop optimization,the Huber loss function is applied for the objective function to reduce the influence of the abnormal quadratic terms.3)The experimental platform of mobile robot is set up,and the experimental verification have been carried out on it.The hardware of the experimental platform is introduced firstly.Based on ROS(Robot Operating System),the data acquisition of sensor and the programming of algorithm software are implemented.Secondly,the parameters for the SLAM algorithm are tuned carefully by experiments.Finally,the experiments about the cumulative error and the mapping precision are carried out in general and weak texture dynamic scenes,and the effectiveness of the proposed approach is confirmed especially in dynamic and weak texture environment.