jackal_3d_slam

This package implements 3D SLAM using RTAB-Map, autonomous navigation using Nav2, and real-time object detection using YOLOv7. In addition, it allows you to perform point cloud processing.

Usage

Before using this package, you must set up the Jackal on ROS 2 Humble. Follow the instructions here to get the Jackal up and running.

Executables

• filter: This runs the point_cloud_processing node, which removes and downsamples noisy point cloud data using PassThrough, RadiusOutlierRemoval, and VoxelGrid filters. The node subscribes to /velodyne_points and publishes sensor_msgs/msg/PointCloud2 messages to /filtered_velodyne_points.
• object_detection.py: This runs YOLOv7 to perform object detection. It is based on the code in a separate repository here.

  Launch Files

• ros2 launch jackal_3d_slam filtered_velodyne.launch.py allows RTAB-Map to use the filtered point cloud to generate an occupancy grid map. The arguments are the following:
  • use_sim_time: Uses simulation clock. Default is false.
  • deskewing: Enables LiDAR deskewing. Default is true.
• ros2 launch jackal_3d_slam jackal_nav.launch.py loads the Nav2 parameters that are modified to work with the Jackal
• ros2 launch jackal_3d_slam jackal_transform.launch.py publishes a static transform between base_link and velodyne, and runs RTAB-Map. The arguments are the following:
  • publish_static_tf: Publishes a static transform between base_link and velodyne. Default is true.
  • use_unfiltered: Launches RTAB-Map with unfiltered point cloud. Default is false.
  • use_filtered: Launches RTAB-Map with filtered point cloud. Default is false.
• ros2 launch jackal_3d_slam object_detection.launch.xml runs YOLOv7 object detection on a RealSense camera
• ros2 launch jackal_3d_slam start_3d_slam.launch.xml starts the Jackal, Velodyne nodes, and Nav2. It runs the point_cloud_processing node depending on the value of the argument below.
  • filter: Determines whether or not to filter the point cloud. Default is false.
• ros2 launch jackal_3d_slam velodyne.launch.py allows RTAB-Map to use the unfiltered point cloud to generate an occupancy grid map. The arguments are the following:
  • use_sim_time: Uses simulation clock. Default is false.
  • deskewing: Enables LiDAR deskewing. Default is true.

Parameters

The following parameters in config/filter_params.yaml can be used to change the filter settings:

• x_filter_min: the minimum value for the PassThrough filter limit in the x direction
• x_filter_max: the maximum value for the PassThrough filter limit in the x direction
• z_filter_min: the minimum value for the PassThrough filter limit in the z direction
• z_filter_max: the maximum value for the PassThrough filter limit in the z direction
• search_radius: the sphere radius used for finding the k-nearest neighbors for the RadiusOutlierRemoval filter
• num_neighbors: the minimum number of neighbors that a point needs to have for the RadiusOutlierRemoval filter
• voxel_leaf_size: the leaf size for the VoxelGrid filter