This release contains the code, released episode-level evaluation records, and reproduction tooling for the single-task ParcelStow evaluation: a matched expert-learner comparison across task execution speeds. It corresponds to the arXiv-v1 manuscript currently in preparation. The arXiv identifier will be added to the README, CITATION.cff, and these notes when assigned.
Reproduce the Released Records
No simulator or GPU is required. The following commands rebuild the success-versus-speed figure, the success table, and the paired-bootstrap interval from the records bundled in data/records/:
python3 -m pip install numpy matplotlib
python3 scripts/reproduce.py envelopeInstallation
The simulation tiers need Isaac Lab (tested with Isaac Sim 5.1). Install the task extension into the Isaac Lab Python environment:
uv pip install -p <isaaclab-venv>/bin/python -e source/parcelstowLightweight Tests
The pure geometry tests run without a simulator:
python3 -m pytest tests/ -qSimulator-backed physics tests require Isaac Lab: python -m pytest tests/ --isaac -q.