Further Exploration
Once you're comfortable with setup, teleoperation, and dataset collection, you can dive deeper into more advanced and creative workflows using LeRobot.
Suggested Experiments
- Try collecting demonstrations for tasks beyond block picking (e.g., stacking, sorting).
- Explore use of dynamic grippers or more complex manipulators.
- Introduce distractors or obstacles and test policy robustness.
Multimodal Data Collection
- Add more camera angles (top-down, wrist-mounted) and use them in training.
Alternative Leader Inputs
- Try joystick control or keyboard teleoperation.
- Use other sensor-based arms or even VR input for richer demonstrations.
Train Your Own Policy
- Swap in different learning algorithms (BC, Dagger, etc.).
- Run ablation studies on camera resolutions, frame rate, or policy architecture.
Contribute Back
- Publish your datasets to the Hugging Face Hub.
- Share your trained models and policies.
- Document hardware modifications for others.