florence-xtrainer2-pick-bottle-uniform

ReTime / xTrainer2 — pick_bottle (absolute EEF, uniform)

Suite index: RainyBot/ReTime-xtrainer2

StarVLA Florence policy on xTrainer2 pick_bottle, absolute bimanual EEF, 10 Hz, ~60 epochs.
Inference bundle (consolidated DeepSpeed ZeRO-3 .pt), not a transformers checkpoint.

Contents

config.yaml
dataset_statistics.json
checkpoints/
  steps_5910.pt    # ~10 ep
  steps_11820.pt   # ~20 ep
  steps_17730.pt   # ~30 ep
  steps_23640.pt   # ~40 ep
  steps_29550.pt   # ~50 ep
  steps_35460.pt   # ~60 ep (recommended)

Requirements (not in this repo)

• microsoft/Florence-2-large (or local path containing florence)
• StarVLA / OPTR with xTrainer2 DataConfig
• pytorch3d

Action / state

• Absolute EEF (not joints / not delta)
• Wire 16-D xyzw; model 20-D (quat → rot6d)
• Cameras: cam_high, cam_left_wrist, cam_right_wrist
• Instructions (exact strings):
  • Pick up the pink primer bottle and place it in the square dish.
  • Pick up the pink dropper and place it in the square dish.
  • Pick up the purple eyebrow gel tube and place it in the square dish.
  • Pick up the silver eyebrow gel tube and place it in the square dish.

Serve

export FLORENCE_MODEL_PATH=/path/to/Florence-2-large
bash examples/Xtrainer2/train_files/run_xtrainer2_policy_server.sh \
  /path/to/this-repo/checkpoints/steps_35460.pt \
  0 \
  10093

Sibling models

• Relative + aug: pick-bottle-rel-aug-uniform
• pack_phone baseline: pack-phone-uniform
• pack_phone residual: pack-phone-residual