D. Reproduce
| Level |
Needs |
What it reproduces |
| 1. Tables |
Python 3.9+, no packages |
Every table in Results from data/droid_episodes.csv |
| 2. Statistics |
the same |
Power table and Wilson bounds (seen_vs_followed.stats) |
| 3. Backbone read test |
one GPU, transformers ≥ 5, the two backbones from the Hugging Face Hub |
droid/backbone_read.py (word recall of each policy's own backbone on the cards) |
| 4. Scenes |
Isaac Sim 5, NVIDIA sim-evals |
droid/build_scenes.py builds the card scenes; droid/render_scene.py renders them at rest |
| 5. Policy runs |
Level 4 plus a policy server per model |
Not included in this prototype (the servers hold machine-specific paths) |
pip install -e .[test]
python -m pytest
python -m seen_vs_followed.report data/droid_episodes.csv
python scripts/gen_docs.py --check