Ground Segmentation ROS 2 Node (GSeg3D)

This package implements a ROS 2 node that wraps the GSeg3D ground segmentation algorithm and exposes it as a real-time perception component for robotic systems.

The node subscribes to LiDAR point clouds (optionally synchronized with IMU data), performs two-phase grid-based ground segmentation, and publishes ground and obstacle point clouds for downstream navigation and perception modules.

You can watch a video of GSeg3D's segmentation results on various datasets at the following link.

Overview

The node:

• Receives a sensor_msgs/PointCloud2 input
• Optionally fuses IMU orientation for gravity alignment
• Applies GSeg3D Phase I (coarse) and Phase II (refinement) segmentation
• Publishes:
  • Ground points
  • Obstacle (non-ground) points
  • Raw filtered input cloud

Subscribed Topics

Topic	Type	Description
/ground_segmentation/input_pointcloud	sensor_msgs/msg/PointCloud2	Input LiDAR point cloud
/ground_segmentation/input_imu	sensor_msgs/msg/Imu	IMU orientation (optional)

IMU synchronization is enabled via the parameter use_imu_orientation.

Published Topics

Topic	Type	Description
/ground_segmentation/ground_points	sensor_msgs/msg/PointCloud2	Segmented ground points
/ground_segmentation/obstacle_points	sensor_msgs/msg/PointCloud2	Segmented non-ground points
/ground_segmentation/raw_points	sensor_msgs/msg/PointCloud2	Filtered input cloud

Processing Pipeline

1. Input Filtering

   • ROI cropping (min/max X, Y, Z)
   • Optional voxel downsampling

2. Synthetic Ground Injection

   • Injects a ground seed beneath the robot
   • Ensures reliable ground region initialization

3. Frame Alignment

   • Transforms point cloud into robot_frame
   • Uses TF2 for frame lookup
   • Uses IMU data (if enabled) to compute gravity-aligned orientation

4. Two-Phase Ground Segmentation

   • Phase I: Coarse grid segmentation (large Z resolution)
   • Phase II: Fine grid refinement (small Z resolution)

5. Post-Processing

   • Radius-based filtering around the robot
   • Final ground / obstacle separation

6. Publishing

   • Results published as ROS 2 PointCloud2 messages

Parameters (Key)

Parameter	Description
robot_frame	Target frame for segmentation
use_imu_orientation	Enable IMU-based gravity alignment
cellSizeX, cellSizeY, cellSizeZ	Grid resolution (Phase I) (m)
cellSizeZPhase2	Grid resolution in z (Phase II) (m)
slopeThresholdDegrees	Max slope for ground (degrees). Controls which planes are considered drivable.
groundInlierThreshold	Plane fitting inlier threshold (m). Controls how far points may lie from the estimated plane.
centroidSearchRadius	KD-tree expansion radius (m)
lidar_to_ground	Lidar to ground distance (m)
transform_tolerance	Tolerance for fetching transforms from TF (sec)
show_benchmark	Enable precision/recall evaluation
maxGroundHeightDeviation	Controls allowed centroid height difference in meters during phase 2 expansion.

Usage Instructions

System Requirements

OS: Ubuntu 22.04, Ubuntu 24.04

ROS2: Humble, Jazzy

Prerequisite

Before building the ROS 2 wrapper, install dependencies and clone the core ground_segmentation library into your ROS 2 workspace:

sudo apt update
sudo apt install cmake libpcl-dev libeigen3-dev libgtest-dev libnanoflann-dev openjdk-17-jre

mkdir -p ~/ros2_ws/src && cd ~/ros2_ws/src
git clone git@github.com:dfki-ric/ground_segmentation.git

Build Instructions

Clone the wrapper into your ROS 2 workspace src folder and build it:

cd ~/ros2_ws/src
git clone git@github.com:dfki-ric/ground_segmentation_ros2.git
cd ~/ros2_ws && source /opt/ros/YOUR_ROS_DISTRO/setup.bash
colcon build --packages-up-to ground_segmentation_ros2 --cmake-args -DCMAKE_BUILD_TYPE=RELEASE
source install/setup.bash

Running the ROS 2 Node

After building the package, launch the ground segmentation node using the provided ROS 2 launch file.

Note:

1. Launch argument imu_topic is optional and is used only when parameter use_imu_orientation is true.
2. If launch file crashes due to a jwt or jvm error then apply fix mentioned here

ros2 launch ground_segmentation_ros2 ground_segmentation.launch.py \
  pointcloud_topic:=<POINTCLOUD_TOPIC> \
  imu_topic:=<IMU_TOPIC> \
  sim:=<true/false>

Replace <POINTCLOUD_TOPIC> and <IMU_TOPIC> with the topics published by your LiDAR and IMU drivers.

sim (boolean) Controls whether simulation time is used.

Behavior:

• true → uses /clock (simulation or bag playback)
• false → uses system wall-clock time

Example:

sim:=true

Node Configuration

• Parameters are loaded from:

  ground_segmentation_ros2/config/parameters.yaml
  

• Topic remapping is handled at launch time, allowing the node to remain independent of sensor-specific topic names.

Runtime Note: libjawt.so

On some systems, the node may fail to start with:

error while loading shared libraries: libjawt.so: cannot open shared object file

FIX

Ensure the JVM libraries are discoverable

Option A (recommended for local development):

export JAVA_HOME=$(dirname $(dirname $(readlink -f $(which javac))))
export LD_LIBRARY_PATH=$JAVA_HOME/lib:$JAVA_HOME/lib/server:$LD_LIBRARY_PATH

Option B (system-wide configuration):

export JAVA_HOME=$(dirname $(dirname $(readlink -f $(which javac))))
echo "$JAVA_HOME/lib" | sudo tee /etc/ld.so.conf.d/java.conf
echo "$JAVA_HOME/lib/server" | sudo tee -a /etc/ld.so.conf.d/java.conf
sudo ldconfig

Benchmarking Mode

When show_benchmark=true:

• Runtime statistics are printed to console

Intended Use

• Real-time ground segmentation for mobile robots
• Safety-critical navigation and obstacle detection
• Traversability analysis
• Research

Notes

• Requires a valid TF tree between sensor, IMU, and robot frames
• Designed for CPU real-time execution

License

BSD-3 Clause License.

Citation

If you use this work, please cite:

Muhammad Haider Khan Lodhi and Christoph Hertzberg, "GSeg3D: A High-Precision Grid-Based Algorithm for Safety-Critical Ground Segmentation in LiDAR Point Clouds" in 2025 7th International Conference on Robotics and Computer Vision (ICRCV), pp. 119-126, 2025. doi: 10.1109/ICRCV67407.2025.11349133

© DFKI Robotics Innovation Center