Yuezhan Tao* · Dexter Ong* · Fernando Cladera · Jason Hughes · Camillo J. Taylor · Pratik Chaudhari · Vijay Kumar
* Equal contribution
HALO is a language-conditioned autonomous aerial exploration and navigation system designed for overhead monocular UAV operation. The system enables a robot to explore unknown environments, build semantic maps, and search for user-specified targets using natural language prompts.
This repository provides the official release, including simulation tools, ROS integration, configuration files, and deployment utilities.
This release includes:
- Source code
- ROS launch files
- Configuration files
- Utility scripts
- tmux session scripts for simulation and robot deployment
- Dockerfiles
- Unity simulation binaries
This repository has been tested with:
- Ubuntu 22.04
- ROS 2 Humble
We provide Docker scripts for a quick simulation setup.
./build_sim.sh./run_sim.shPlease download the Unity simulation binaries from the following link:
After downloading, unzip the binaries and copy them into the Docker container:
docker cp aerial_sim_binaries/ air_sem_exp:/root/ros2_ws/src/
The provided tmux script launches the full simulation stack using tmuxp.
cd ros2_ws/src/air_sem_explorer/tmux
tmuxp load sim.yamlThis launches:
- Zenoh router
- Unity simulator
- Planner node
- Geometric mapper node
- Semantic mapper node
- Simulation tracker node
To trigger exploration, go to the Trigger tmux tab and press Enter.
To shut down all processes, go to the final kill tmux tab and press Enter.
Edit:
air_sem_explorer/config/map_common.yamlEdit:
air_sem_gridmap/config/config.yamlUse:
ros2 service call /set_prompt air_sem_gridmap_interfaces/srv/SetPrompt "{prompt: '<prompt1,prompt2>'}"When using the tmux script, a GUI window will also appear to allow you to set the task prompt.
We thank the authors of the following repositories for their open-source code:
If you find our paper or code useful, please consider citing:
@article{tao2025halo,
title={HALO: High-Altitude Language-Conditioned Monocular Aerial Exploration and Navigation},
author={Tao, Yuezhan and Ong, Dexter and Cladera, Fernando and Hughes, Jason and Taylor, Camillo J and Chaudhari, Pratik and Kumar, Vijay},
journal={arXiv preprint arXiv:2511.17497},
year={2025}
}
