This package is for 3D mapping and localization using NDT (normal-distributions transform). Includes a simple loop closure feature without any total optimization. Multiprocessing NDT is available using ndt_omp.
This package has been developed and tested under ROS noetic on Ubuntu 20.04.
ndt_mapper_demo_video.mp4
- sensor_msgs/PointCloud2 : 3D LiDAR scan data
- tf2_msgs/TFMessage : transform info between robot's origin and lidar
- nav_msgs/Odometry : (optional) can be used for initial guess of next scan
Clone ndt_omp in your ROS workspace for OpenMP acceleration of NDT calculations.
git clone https://github.com/koide3/ndt_ompAfter that, please clean and re-build this package (or your entire workspace) to apply changes.
OpenMP will automatically be used after installation and rebuild.
-
Launch mapping nodes.
roslaunch ndt_localizer ndt_mapper.launch
-
Save maps.
roslaunch ndt_localizer map_saver.launch
-
launch arguments for
ndt_mapper.launcharg name description map_frame frame name of map base_frame frame name of robot's origin for odometry lidar_frame frame name of lidar points_topic topic name to subscribe lidar points odom_topic topic name to subscribe odometry map_topic topic name to publish created map map_publish_interval minimal interval to publish the full map [sec] pcd_dense_option sets scan pcd's "is_dense" option to this bool min_scan_range minimal scan range [m] max_scan_range maximum scan range [m] voxel_leaf_size voxel leaf size for scan [m] ndt_max_iterations maximum iteration counts for optimization [-] ndt_resolution voxel leaf size for target map [m] ndt_step_size maximum step length of newton line search [m] ndt_transformation_epsilon maximum translation squared difference for the optimization to be considered as having converged [m^2] prediction_method method to predict pose for next scan; zero, linear or odom translation_error_tolerance maximum limit of translation error between the predicted pose and the NDT optimized pose [m] rotation_error_tolerance maximum limit of rotation error between the predicted pose and the NDT optimized pose [deg] map_add_shift minimal shift length to add current scan to the map [m] submap_scan_size number of scans to include in a single submap [-] submap_include_distance submaps within this distance will be added to the target map for localization [m] submap_connect_distance submaps within this distance will be connected in the pose graph [m] use_loop_closure true to apply loop closure initial_guess_resolution distance between initial guesses when attempting loop closure [m] initial_guess_count number of initial guesses prepared for each direction [-] loop_score_limit limit of ndt fitness score to attempt closing a loop [-] loop_confirmation_translation_tolerance maximum translation error for the loop closure to be considered as having converged [m] loop_confirmation_rotation_tolerance maximum rotation error for the loop closure to be considered as having converged [deg] save_uncompressed_map true to save the full map to a pcd file with no binary compression save_submaps true to save each submap as a single pcd file openmp_thread_num number of threads when using OpenMP [-] openmp_neighbor_search_method method to get neighbor voxels for ndt gradient calculation; 0: KDTREE, 1: DIRECT26, 2: DIRECT7, 3: DIRECT1. For further information, check ndt_omp. use_rviz true to launch rviz -
launch arguments for
map_saver.launcharg name description maps_directory path of directory to save the maps map_name name of the map to be saved
-
Launch localization nodes.
roslaunch ndt_localizer ndt_localizer.launch
-
Set initial pose with
2D Pose Estimatein rviz.
- launch arguments for
ndt_localizer.launcharg name description maps_directory path of directory to find map files map_name name of the map to be used map_frame frame name of map base_frame frame name of robot's origin for odometry lidar_frame frame name of lidar points_topic topic name to subscribe lidar points odom_topic topic name to subscribe odometry pcd_dense_option sets scan pcd's "is_dense" option to this bool min_scan_range minimal scan range [m] max_scan_range maximum scan range [m] voxel_leaf_size voxel leaf size for scan [m] ndt_max_iterations maximum iteration counts for optimization [-] ndt_resolution voxel leaf size for target map [m] ndt_step_size maximum step length of newton line search [m] ndt_transformation_epsilon maximum translation squared difference for the optimization to be considered as having converged [m^2] prediction_method method to predict pose for next scan; zero, linear or odom translation_error_tolerance maximum limit of translation error between the predicted pose and the NDT optimized pose [m] rotation_error_tolerance maximum limit of rotation error between the predicted pose and the NDT optimized pose [deg] use_submaps true to use submaps for localization submap_include_distance submaps within this distance will be added to the target map for localization [m] submap_update_shift minimal shift length for the target map to be updated [m] openmp_thread_num number of threads when using OpenMP [-] openmp_neighbor_search_method method to get neighbor voxels for ndt gradient calculation; 0: KDTREE, 1: DIRECT26, 2: DIRECT7, 3: DIRECT1. For further information, check ndt_omp. use_rviz true to launch rviz