| # 🤖 Robot Data Collection Guide | |
| A practical guide for collecting clean, diverse, and meaningful datasets to train and evaluate robot models. | |
| --- | |
| ## 📌 Why This Matters | |
| High-quality data is the foundation for: | |
| - Behavior cloning | |
| - Multi-robot coordination | |
| - Sim-to-real transfer | |
| - Sensor fusion and modeling | |
| --- | |
| ## 📂 What's Inside | |
| - ✅ Types of data to collect (images, joint states, LiDAR, IMU) | |
| - ✅ How to record multi-modal data (ROS 2 bag files) | |
| - ✅ Labeling methods (manual, semi-auto) | |
| - ✅ Data cleaning and preprocessing | |
| - ✅ Storage & formats: `.db3`, `.csv`, `JSON` | |
| --- | |
| ## 🛠 Tools You Can Use | |
| | Tool | Purpose | | |
| |-----------------|----------------------------------| | |
| | ROS 2 | Real-time data collection | | |
| | `rosbag2_py` | Python access to .db3 files | | |
| | DB Browser | Inspect SQLite `.db3` bags | | |
| | OpenCV / PIL | Preprocess visual data | | |
| | Pandas | Work with CSVs | | |
| | Hugging Face | Store models/datasets online | | |
| --- | |
| ## 📦 Example Commands | |
| ```bash | |
| # Record robot states + camera stream | |
| ros2 bag record /joint_states /camera/image_raw /tf | |