nvblox_fastlio_bridge

GPU-ready ROS 2 bridge that feeds FAST-LIO state estimation directly into NVBlox for real-time TSDF, mesh, and ESDF map generation on Jetson Orin.

Overview

This package provides a high-performance bridge between FAST-LIO (LiDAR-inertial odometry) and nvblox (GPU-accelerated volumetric mapping). It consumes the registered point clouds and odometry from FAST-LIO to generate dense 3D reconstructions in real-time on NVIDIA Jetson platforms.

Key Features

• Seamless Integration: Directly subscribes to /cloud_registered and /Odometry from FAST-LIO.
• Jetson Optimized: Tuned for NVIDIA Jetson Orin (SM 8.7) with CUDA 12.8.
• Real-time Reconstruction: Generates TSDF, ESDF, and Mesh layers at LiDAR frame rates (10-20 Hz).
• Rich Visualization: Pre-configured RViz setup for viewing meshes, voxel grids, and distance fields.
• Voxel Export: Includes functionality to export the reconstructed map to PCD files.

Repository Layout

Path	Purpose
launch/	ROS 2 launch files (nvblox_fastlio.launch.py).
src/	Core bridge source code.
config/	Configuration files for nvblox and FAST-LIO.
rviz/	Visualization configuration (nvblox_mapper.rviz).

Getting Started

Prerequisites

• Hardware: NVIDIA Jetson Orin (or other CUDA-capable GPU).
• OS: Ubuntu 22.04 (ROS 2 Humble).
• Dependencies:
  • ROS 2 Humble
  • CUDA Toolkit (12.8+ recommended for Orin)
  • nvblox (installed and built)
  • FAST_LIO (GPU accelerated version)

Installation

1. Clone the repository into your ROS 2 workspace:

   cd ~/ros2_ws/src
   git clone git@github.com:OmerMersin/fastlio_nvblox_mapper.git nvblox_fastlio_bridge

2. Build the package:

   cd ~/ros2_ws
   colcon build --packages-select nvblox_fastlio_bridge
   source install/setup.bash

Usage

Launching the Mapper

To launch FAST-LIO and the nvblox bridge together with RViz visualization:

ros2 launch nvblox_fastlio_bridge nvblox_fastlio.launch.py \
  fastlio_config:=/path/to/fast_lio/config/ouster32.yaml \
  bridge_config:=/path/to/nvblox_fastlio_bridge/config/nvblox_params.yaml

Arguments:

• fastlio_config: Absolute path to your FAST-LIO sensor configuration file (default: fast_lio/config/ouster32.yaml).
• bridge_config: Path to the nvblox parameter file (default: config/nvblox_params.yaml).
• launch_rviz: Set to true or false to enable/disable RViz (default: true).

Example Command

ros2 launch nvblox_fastlio_bridge nvblox_fastlio.launch.py \
  fastlio_config:=/home/orin/LIONav/ros2_ws/src/FAST_LIO/config/ouster32.yaml \
  bridge_config:=/home/orin/LIONav/ros2_ws/src/nvblox_fastlio_bridge/config/nvblox_params.yaml

Configuration

nvblox Parameters

The mapping behavior is controlled by config/nvblox_params.yaml. Key parameters include:

• voxel_size: Resolution of the map (default: 0.10m).
• truncation_distance_vox: Truncation distance in voxel units.
• esdf_update_rate_hz: Frequency of ESDF updates.

FAST-LIO Parameters

Refer to the FAST-LIO repository for details on sensor configuration (LiDAR type, extrinsics, etc.).

Visualization

The launch file automatically opens RViz with a pre-configured view (rviz/nvblox_mapper.rviz) showing:

• Point Cloud: The raw registered cloud from FAST-LIO.
• Mesh: The reconstructed mesh from nvblox.
• Odometry: The path and pose of the sensor. | max_integration_distance | 15.0 | Clip lidar rays beyond this range. | | pointcloud_topic | /cloud_registered | FAST-LIO registered cloud input. | | odom_topic | /Odometry | FAST-LIO pose stream used for TSDF poses. | | publish_voxels | true | Toggle voxel CUBE_LIST marker publisher. | | publish_voxel_points | false | Publish sampled voxel point cloud for analytics. | | voxel_export_* | see nvblox_fastlio_node.cpp | Filters for visualization and PCD exports. |

All parameters are declared on the node, so you can override them via launch files, YAML, or ros2 param set at runtime.

Runtime interfaces

• Subscriptions: /cloud_registered (sensor_msgs/PointCloud2), /Odometry (nav_msgs/Odometry).
• Publishers:
  • nvblox/mesh (visualization_msgs/Marker TRIANGLE_LIST)
  • nvblox/voxels (visualization_msgs/Marker CUBE_LIST)
  • nvblox/voxel_points (sensor_msgs/PointCloud2, optional)
  • nvblox/esdf_slice, nvblox/tsdf_slice (slice point clouds)
  • nvblox/occupancy_grid (nav_msgs/OccupancyGrid)
• Services: nvblox/save_voxels (std_srvs/Trigger) → writes filtered TSDF voxels to PCD.

Tips for best results

1. Clock alignment: make sure FAST-LIO and NVBlox run on synchronized clocks (use /use_sim_time or PTP if needed).
2. TF availability: keep map -> body/lidar transforms continuous; the node falls back to raw odometry, but TF avoids drift during temporary dropouts.
3. Denoising toggles: lidar_denoise_*, lidar_smooth_*, and lidar_hole_filling_radius help when Ouster returns sparse walls or Livox has motion blur.
4. Jetson thermals: pin clocks with jetson_clocks and watch GPU temps when integrating dense indoor scans.

License

Released under the MIT License.