SubtPlanner: Structure-Guided Sampling and Branch-Aware Repositioning for UAV Autonomous Subterranean Exploration

✨ Overview

SubtPlanner is a cognitive-map-based hierarchical active planner for UAV autonomous exploration in subterranean environments. It couples short-horizon local exploration with long-term topological memory, allowing the robot to keep exploring efficiently in long corridors, branch-rich tunnels, caves, and dead-end-prone spaces.

The planner is organized around three method concepts:

• Cognitive map: an incremental representation composed of a local exploration graph and a global topological graph.
• Local exploration: structure-guided sampling using passable directions, field-of-view gain, and safety corridors, dead-zone recognition, and path utility selection.
• Global repositioning: dead-zone escape and branch-aware frontier selection with return-path reasoning.

🗞️ Updates

• 2026.09 The manuscript has been submitted to ICRA 2027!

🎞️ Simulation Preview

🗂️ Simulation Assets

This repository includes the lightweight urban_corridor demo assets for quick testing. Large subterranean simulation environments are based on public DARPA SubT / Open Robotics simulation assets. Please obtain the original large-scale simulation assets from the official sources:

• OSRF SubT repository: https://github.com/osrf/subt
• NTNU-ARL subt_cave_sim: https://github.com/ntnu-arl/subt_cave_sim

⚙️ Quick Start

📦 1. Prerequisites

Required base environment: Ubuntu 20.04, ROS Noetic, GCC 9 / C++17, and Gazebo Classic 11.

sudo apt update
sudo apt install -y \
  build-essential cmake git git-lfs curl wget \
  python3-catkin-tools python3-vcstool python3-rosdep python3-pip \
  libboost-all-dev libeigen3-dev libgoogle-glog-dev libgflags-dev \
  libprotobuf-dev protobuf-compiler libyaml-cpp-dev libsqlite3-dev libzmq3-dev \
  libpcl-dev libopencv-dev qtbase5-dev

git lfs install
gazebo --version
pkg-config --modversion gazebo

📥 2. Create Workspace

mkdir -p ~/SubtPlanner_ws/src
cd ~/SubtPlanner_ws/src
git clone https://github.com/PengSO/SubtPlanner.git

# Download external simulation dependency: rotors_simulator 
cd simulation
git clone https://github.com/ethz-asl/rotors_simulator.git
cd ..

Optional: recover Git LFS files if checkout failed

cd ~/SubtPlanner_ws/src
git lfs pull
git restore --source=HEAD :/

🛠️ 3. Build

cd ~/SubtPlanner_ws
./src/SubtPlanner/scripts/build_subtplanner.sh ~/SubtPlanner_ws

Optional: manual build commands

cd ~/SubtPlanner_ws
source ./src/scripts/setup_gazebo_classic_env.sh
rosdep install --from-paths src --ignore-src -r -y
catkin config --cmake-args -DCMAKE_BUILD_TYPE=Release
catkin build
source devel/setup.bash
./src/SubtPlanner/scripts/check_runtime_env.sh ~/SubtPlanner_ws

Optional: rebuild simulator plugins if the Gazebo environment changed

Use this only when the workspace was previously built under a different Gazebo setup, for example after switching from Gazebo Sim / ROS-GZ back to Gazebo Classic 11.

cd ~/SubtPlanner_ws
./src/SubtPlanner/scripts/rebuild_sim_plugins.sh ~/SubtPlanner_ws

Recommended once after building:

echo 'source ~/SubtPlanner_ws/src/SubtPlanner/scripts/setup_gazebo_classic_env.sh' >> ~/.bashrc
source ~/.bashrc

🚁 4. Run Corridor Demo

cd ~/SubtPlanner_ws
source devel/setup.bash
roslaunch subt_planner urban_corridor.launch

Optional: run with a temporary shell environment

cd ~/SubtPlanner_ws
source ./src/SubtPlanner/scripts/setup_gazebo_classic_env.sh
source devel/setup.bash
roslaunch subt_planner urban_corridor.launch

Optional: headless smoke test

cd ~/SubtPlanner_ws
source devel/setup.bash
roslaunch subt_planner urban_corridor.launch rviz_en:=false

📖 Citation

If this project supports your research, please cite the associated paper:

@inproceedings{subtplanner2027,
  title = {SubtPlanner: Structure-Guided Sampling and Branch-Aware Repositioning for UAV Autonomous Subterranean Exploration},
  author = {Peng, Song and Zhang, Hao and Li, Han and Jiang, Weicheng and Gou, Guohua and Sui, Haigang},
  booktitle = {},
  year = {2027}
}

🙏 Acknowledgements

We thank the following projects for their valuable influence on subterranean exploration research and simulation workflows:

• GBPlanner2
• OmniPlanner
• SuperLoc
• voxblox
• DARPA SubT Challenge / Open Robotics official simulation assets