Datasets:

anonymous-rabench-2026
/

runtime-authority-bench

license: cc-by-4.0
language:
  - en
pretty_name: RA-Bench Runtime Authority Review Artifact
tags:
  - robotics
  - robot-learning
  - benchmark
  - mujoco
  - runtime-verification
  - safety-evaluation
  - reproducibility

RA-Bench Runtime Authority Review Artifact

This anonymous review artifact contains the supplementary material for a NeurIPS 2026 Evaluations and Datasets submission on runtime authority evaluation for learned robot policies in structured-state simulation.

Files

supplementary_material.zip: self-contained benchmark, compact logs, figures, reports, verifier scripts, and reproduction checks.
croissant_metadata.json: Croissant metadata with core fields and Responsible AI metadata.

Scope

The artifact supports structured-state MuJoCo tabletop pre-grasp simulation claims only. It should not be interpreted as evidence for physical robot safety, real camera sensing, real tactile sensing, real audio sensing, ROS2 hardware control, or deployment-ready robot control.

Integrity

supplementary_material.zip SHA256: 3975c65fcfd00a3232e50fa871ebb48431eed9863503f9315ff3066685519e46
croissant_metadata.json SHA256: 752d1d7ade40cdbca068420a3078d52d9d53195780c70296b604703e6d931221

Review Use

Reviewers can download supplementary_material.zip, unpack it, and start from:

python3 supplementary_material/submission_checks/VERIFY_SUPPLEMENT.py
python3 supplementary_material/reproduction_core/RUN_REPRODUCIBILITY_CHECKS.py

The package is intentionally compact and contains selected logs plus aggregate summaries rather than raw full video or audio streams.