🤖 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
# Record robot states + camera stream
ros2 bag record /joint_states /camera/image_raw /tf