FTP-1 v0426 50kstep

ftp1_pretrain_v0426_50kstep is the 50k-step pretrained checkpoint of FTP-1, a general tactile policy for multimodal manipulation with vision, proprioception, language, and optional tactile input.

This checkpoint is mainly intended as a base model for downstream fine-tuning and research use.

FTP-1 is described in the main project materials as being pretrained on approximately 3,000 hours of heterogeneous tactile manipulation data.

Notes

• Type: pretrained checkpoint
• Stage: v0426, 50k steps
• Inputs: image, state, language, optional tactile
• Output: action chunk prediction

Links

• Project homepage: https://ftp1-policy.github.io/
• Open-source repository: https://github.com/michaelyuancb/ftp1-policy