RLinf/RLinf-Pi05-RealWorld-DualFranka-Clean_Desk

Pi0.5 Dual-Franka Clean Desk

Pi0.5 checkpoint used for the dual-Franka clean-desk demo in RPent. It provides learned grasping, handoff and placement actions; dirty/clean sorting decisions are made by the RPent planner.

Download and use

hf download RLinf/RLinf-Pi05-RealWorld-DualFranka-Clean_Desk \
  --revision 4f43adc723dfe98fae806a131ad61999e1f95e4f \
  --local-dir ./models/pi05-dualfranka-clean-desk

export PI05_CHECKPOINT_PATH="$PWD/models/pi05-dualfranka-clean-desk"
export DUAL_FRANKA_REPO_ID=id_0

The checkpoint directory contains:

actor/model_state_dict/full_weights.pt
id_0/norm_stats.json

id_0 identifies the matching normalization statistics, not the Hugging Face model repository. The deployed policy predicts 20-step action chunks.

Follow the RPent dual-Franka guide for hardware setup and execution. This checkpoint was trained for our in-house robot, camera and workspace setup; performance on other setups is not guaranteed. Collect demonstrations and fine-tune for your own setup as needed.