Semantic Segmentation Simulation

Gazebo simulation environment for testing semantic segmentation with Nav2 integration.

Overview

This package provides a complete simulation environment with:

• TurtleBot 4 robot model with RGBD camera
• Baylands world with sidewalk and grass terrain
• Nav2 integration with semantic segmentation costmap layer
• RViz visualization configuration

The semantic segmentation inference is provided by the semantic_segmentation_node package.

Launch Files

nav2_segmentation_launch.py

Main launch file that starts:

• Nav2 navigation stack with semantic segmentation costmap layer
• Semantic segmentation inference node (ONNX-based)
• RViz (optional)

ros2 launch semantic_segmentation_sim nav2_segmentation_launch.py rviz:=true

simulation_launch.py

Starts the Gazebo simulation environment:

• Gazebo with Baylands world
• TurtleBot 4 robot spawn
• Static transforms

ros2 launch semantic_segmentation_sim simulation_launch.py headless:=false

segmentation_simulation_launch.py

Complete launch file that starts both simulation and navigation:

ros2 launch semantic_segmentation_sim segmentation_simulation_launch.py

Arguments:

• use_sim_gui - Launch Gazebo with GUI (default: true)
• use_rviz - Launch RViz visualization (default: true)

Example:

# Launch without GUI and without RViz
ros2 launch semantic_segmentation_sim segmentation_simulation_launch.py use_sim_gui:=false use_rviz:=false

Configuration

• config/nav2_params.yaml - Nav2 parameters with semantic segmentation layer config
• config/segmentation_rviz_config.rviz - RViz configuration for visualization
• worlds/baylands.sdf - Gazebo world file

Semantic Segmentation Integration

The simulation uses a custom Nav2 costmap layer (semantic_segmentation_layer) that:

• Subscribes to segmentation masks, confidence, and point clouds
• Projects segmentation onto the costmap
• Assigns costs based on terrain class (sidewalk = traversable, grass = danger)

See config/nav2_params.yaml for detailed layer configuration.