Does nvblox support loop closure?

Robotics & Edge Computing Robotics - Isaac Isaac ROS
,
1

Hi there,

I’m trying to take a side-by-side stereo video feed, run through ESS Disparity and VSLAM nodes, and then feed the outputs into NVblox to get a 3D reconstruction in realtime.

I’ve noticed that when the VSLAM node successfully performs loop closure during SLAM, only the robot’s position updates and the nvblox-generated mesh does not adjust from the loop closure / realign.

Does Isaac ROS Nvblox support loop closure? If so, how can I achieve this?

Below is how I am launching the VSLAM and nvblox nodes:

    visual_slam_node = ComposableNode(
        name='visual_slam_node',
        package='isaac_ros_visual_slam',
        plugin='nvidia::isaac_ros::visual_slam::VisualSlamNode',
        remappings=[('stereo_camera/left/image', '/left/image_raw_grayscale'),
                    ('stereo_camera/left/camera_info', '/left/camera_info'),
                    ('stereo_camera/right/image', '/right/image_raw_grayscale'),
                    ('stereo_camera/right/camera_info', '/right/camera_info')],
        parameters=[{
                    'enable_rectified_pose': True,
                    'denoise_input_images': False,
                    'rectified_images': False,
                    'img_jitter_threshold_ms': 110.00, # default is 33.33 (30fps). Input to VSLAM is ~23 fps for gscam mode. On Orin, ~10fps
                    'enable_slam_visualization': True,
                    'enable_observations_view': True,
                    'enable_landmarks_view': True,
                    "enable_localization_n_mapping": True,
                    "enable_imu_fusion": False,
                    'map_frame': 'map',
                    'odom_frame': 'odom',
                    'base_frame': 'base_link',
                    "input_left_camera_frame": "stereocamera_left_frame", 
                    "input_imu_frame": "imu",
                    }],
    )

    # Nvblox node
    nvblox_node = ComposableNode(
        name='nvblox_node',
        package='nvblox_ros',
        plugin='nvblox::NvbloxNode',
        parameters = ["/workspaces_2/isaac_ros-dev/src/custom_package/config/nvblox_params.yaml"],
        remappings=[
            ('color/image', '/left/image_rect_resize'),
            ('color/camera_info', '/left/camera_info_rect_resize'),
            ('depth/camera_info', '/left/camera_info_rect_resize'),
            ('depth/image', '/depth'),
            # 'transform' and 'pose' topics not required, since params.yaml has use_tf_transforms=true
        ]
    )

and the nvblox_params.yaml:

/**:
  ros__parameters:
    # miscellaneous
    global_frame: "map"
    voxel_size: 0.02 # voxel size in cm
    use_tf_transforms: true
    # multi mapper
    mapping_type: "static_tsdf"  # ["static_tsdf", "static_occupancy"]
    connected_mask_component_size_threshold: 2000
    # esdf settings
    compute_esdf: false
    publish_esdf_distance_slice: false
    # mesh settings
    compute_mesh: true
    mesh_update_rate_hz: 30.0
    # color settings
    use_color: true
    max_color_update_hz: 30.0
    # depth settings
    use_depth: true
    max_depth_update_hz: 30.0
    # lidar settings
    use_lidar: false
    # Input queues
    max_poll_rate_hz: 30.0
    maximum_sensor_message_queue_length: 30
    # Map clearing settings
    map_clearing_radius_m: -1.0 # no map clearing if < 0.0
    map_clearing_frame_id: "base_link"
    # QoS settings
    depth_qos: "SENSOR_DATA"
    color_qos: "SENSOR_DATA"
    
    static_mapper:
      # projective integrator (tsdf/color/occupancy)
      projective_integrator_max_integration_distance_m: 2.0
      projective_integrator_truncation_distance_vox: 50.0
      weighting_mode: "inverse_square"
      projective_integrator_max_weight: 1000.0 
      # mesh integrator
      mesh_integrator_min_weight: 100.0
      # tsdf decay integrator
      tsdf_decay_factor: 0.95
2

Hi @HiddenNetwork

Nvblox is designed to do local mapping, and we mostly use the odom frame as our global_frame and do not handle loop closure.

We recommend setting the map_clearing_radius_m param to clear the map 10 m radius of the robot and, by doing so, constrain the map to a local area around the robot.

1 Like
3

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.