Habitat Data Collector is a standalone simulation application built on top of Habitat-Sim and Habitat-Lab. It also allows researchers to customize their own object settings within Habitat-Sim and create dynamic scenes. It supports scene loading, object insertion/removal, ROS2 output (pose, RGBD stream), and data recording. This tool is designed for collecting and testing datasets for robot perception, navigation, and mapping tasks in both static and dynamic scenes.
- Environment Setup
- Dataset Setup
- Configuration Guide
- Run the Collector
- User Guide
- Project Structure
- Acknowledgment
🖥️ This setup is tested on Ubuntu 22.04 with Python 3.10.
git clone --recurse-submodules https://github.com/Eku127/habitat-data-collector.git
cd habitat-data-collectorconda env create -f environment.yml
conda activate habitat_data_collectorThis step will take some time as it compiles Habitat-Sim from source. Habitat cannot be installed by conda in Python 3.10, so it must be built manually.
bash scripts/install_habitat.shDuring compiling with habitat-sim, if having error with OgenGL, like
Could NOT find OpenGLand errors with compilingzlib_external, install the required libs by:sudo apt install libgl1-mesa-dev libglu1-mesa-dev freeglut3-dev zlib1g-dev
Before running the tool, please follow the dataset setup guide to prepare the required datasets.
For a detailed explanation of configuration options and structure, please refer to the Configuration Reference. Setting up correct configs is crucial for running this tool.
Run the main simulation from the root directory:
python -m habitat_data_collector.mainBy default, it uses the configuration file at: config/habitat_data_collector.yaml. For config details, refer to the Config Reference.
If you want to receive and send ROS2 topic outputs or record ROS2 bags:
- Install ROS2 Humble following the official guide.
- Source the ROS2 environment before running the collector:
source /opt/ros/humble/setup.bash # or setup.zshOnce sourced, the simulator will publish data to ROS2 topics. You can record them by enabling the ROS recording configuration in config/habitat_data_collector.yaml. See the ROS Integration Documentation for topic configuration and ROS2-to-ROS1 bridge setup.
Once the simulator launches successfully, refer to the Usage Guide to learn how to:
- Move the camera and explore the scene
- Add, place, grab, and delete objects
- Start and stop recording (raw data + ROS2 bag)
- Save and reload a scene configuration
The guide includes visual previews and terminal output samples for better understanding.
habitat-data-collector/
├── habitat_data_collector/ # Main application code
│ ├── main.py
│ └── utils/
├── config/ # YAML configuration files
├── 3rdparty/ # Git submodules: habitat-sim & habitat-lab
├── documents/ # Markdown documentation and media
├── scripts/ # Helper scripts (e.g. build, setup)
├── environment.yml # Conda environment spec
└── README.md
- ROS2 Humble must be installed and sourced before using ROS features.
- Configurations are handled with OmegaConf and Hydra.
- All paths, topics, and behaviors are configured in
config/habitat_data_collector.yaml.
This project builds on the outstanding work of:
We thank the authors and contributors of these projects for making them open-source and actively maintained.
This project is also inspired by the data collection pipeline from VLMaps, and we are grateful to the authors of both HOVSG and VLMaps for their contributions.
Special thanks to @TOM-Huang and @aclegg3 for valuable advice and support during development.
MIT License