SPOMP: Semantic Panoramic Online Mapping and Planning

This is the public code repository for our work SPOMP. Here are all the submodules needed for the stack.

Video

How to run the code?

We provide Docker images that can be used to run this code:

# Clone and run the master docker image
git clone https://github.com/KumarRobotics/dcist_master.git
cd dcist_master
./run.bash dcist-master

# Clone the repo and build
cd user_ws
mkdir src
git clone --recursive -j8 https://github.com/KumarRobotics/spomp-system.git src/spomp-system
catkin config --extend ~/dcist_ws/devel && catkin build -DCMAKE_BUILD_TYPE=Release

Launch files for air and ground robots can be found inside semantics_manager.

Main modules

• semantics_manager: Starting point for launch files and configuration management
• ROFL: LiDAR odometry
• SPOMP: Autonomy stack for UGVs. Handles terrain analysis and global and local planning.
• ASOOM: Aerial orthomapper
• top_down_renderer: Crossview localizer
• air_router: High-level planner for UAV
• MOCHA: Distributed, opportunistic communication framework

Dependencies

• grid_map: Fork of grid_map with resizing and compression capabilities
• orbslam3_ros: ROS wrapper for ORBSLAM3
• erfnet_pytorch_ros: ROS wrapper for erfnet (for aerial image segmentation)
• rangenet_inf: ROS wrapper for RangeNet++ (for depth panorama segmentation)
• ouster_decoder: Ouster LiDAR driver (needed for ROFL)
• ORB_SLAM3: Fork of ORB_SLAM3 with loop closure deactivated.

Tools

These are not needed to run the stack, but may be useful.

• SILL: Depth panorama/LiDAR labelling tool
• rviz_textured_quads Useful tool for visualizing map images in rviz