pedroset/GR00T-N1.6-G1-PnPAppleToPlate-CW

This model has moved!

This repository has been migrated to the CloudWalk Research organization.

New location: cloudwalk-research/GR00T-N1.6-G1-PnPAppleToPlate

Quick update

# Old
model_path = "pedroset/GR00T-N1.6-G1-PnPAppleToPlate-CW"

# New
model_path = "cloudwalk-research/GR00T-N1.6-G1-PnPAppleToPlate"

Why the move?

This model was originally uploaded to my personal account during the research phase. Now that the work is published and gaining traction (90+ downloads in the first month!), we've moved it to our organization's official account for long-term maintenance.

The model weights, configs, and all files remain identical. Future updates (new checkpoints, improved results) will be published at the new location.

What is this model?

Community fine-tune of GR00T N1.6 on Unitree G1 for the PnPAppleToPlate task, achieving 50% closed-loop success rate (5/10 episodes). This was the first community reproduction of NVIDIA's reported fine-tuning results, addressing issues reported in #574, #470, #543, and #558.

Full documentation, reproduction instructions, and training details are available at the new location.

Links

• Model: cloudwalk-research/GR00T-N1.6-G1-PnPAppleToPlate
• Organization: cloudwalk-research
• Fine-tuned by: CloudWalk as part of the SONIC project
• Base model: nvidia/GR00T-N1.6-3B