DreamFAR: Real-Time Robot-Object Interaction Model
1 min readupdated
Developed by Amazon-FAR, this model is designed for bimanual robot manipulation and real-time dynamics prediction using teleoperation data. It includes dynamics checkpoints, an image tokenizer, and test scenes. The solution is intended for robotics researchers and engineers working with simulation and manipulator control.
Key features
- Acts as a world model for bimanual manipulation
- Includes checkpoints, a tokenizer, and test scenes
- Trained on robotic teleoperation data
- Uses bf16 weights for inference
What you can do
- Run the interactive demo from the project repository
- Apply it to simulate robot actions in an environment
- Use the attached test scenes to evaluate dynamics
hf download Amazon-FAR/dreamfar --local-dir dreamfar_assets
(cd dreamfar_assets && sha256sum -c SHA256SUMS && cd scenes && sha256sum -c SHA256SUMS)Read more
DetailsThe model is distributed under the CC BY-NC 4.0 license.
Includes a 386 MB tokenizer and a 4.4 GB dynamics model.
Requires downloading via the official Hugging Face CLI tool.










Comments