Skip to main content

​
Overview

The hazard_detection_node processes depth camera point cloud data to detect hazards such as boulders, craters, and other obstacles in the robot’s environment. It uses Open3D for point cloud processing and implements a plane segmentation algorithm to identify obstacles that deviate from the ground plane. The node performs the following processing pipeline:
  1. Filters invalid points (NaN and infinite values)
  2. Downsamples the point cloud using voxel filtering
  3. Removes statistical outliers
  4. Segments the ground plane using RANSAC
  5. Publishes remaining points as detected obstacles
These obstacle points can be integrated into the navigation costmap for path planning and collision avoidance.

​
Parameters

This node does not declare any configurable ROS parameters. Processing parameters are hardcoded:
  • Voxel downsampling size: 0.05 m
  • Statistical outlier removal: 20 neighbors, 2.0 std deviation ratio
  • Plane segmentation distance threshold: 0.05 m
  • RANSAC iterations: 1000

​
Published Topics

​
/hazards/front

Type: sensor_msgs/msg/PointCloud2 Description: Filtered point cloud containing detected obstacles after ground plane removal. Points are published in the same frame as the input camera data. This topic can be consumed by costmap layers or visualization tools.

​
Subscribed Topics

​
/camera_front/points

Type: sensor_msgs/msg/PointCloud2 Description: Raw point cloud data from the front-facing depth camera. Expected to contain XYZ fields. The node filters out NaN and infinite values during processing.

​
Action Servers

This node does not provide any action servers.

​
Launch Command

ros2 run lunabot_perception hazard_detection_node
With remapped topics: 
ros2 run lunabot_perception hazard_detection_node \
  --ros-args \
  -r /camera_front/points:=/custom/depth/points \
  -r /hazards/front:=/custom/hazards

Build docs developers (and LLMs) love

Get started for freeTalk to us