Jichao Zhang commited on
Commit
8d3f4b2
·
verified ·
1 Parent(s): 531d72b

Upload cleaned NeurIPS supplement artifacts

Browse files
README.md CHANGED
@@ -28,8 +28,8 @@ The artifact supports structured-state MuJoCo tabletop pre-grasp simulation clai
28
 
29
  ## Integrity
30
 
31
- - `supplementary_material.zip` SHA256: `37d1c14fa46dc0325e94fdb077d239524d9e3ebdd0e34ea52c93cd9c42719f62`
32
- - `croissant_metadata.json` SHA256: `0f144e5e8f0cbb42ce0787f6cdf6b953405ecec61af4d65a934d67485768394d`
33
 
34
  ## Review Use
35
 
 
28
 
29
  ## Integrity
30
 
31
+ - `supplementary_material.zip` SHA256: `1610e821b82fa73b2639b4e8200067ee989625153b12115cb99b9d4384c7878b`
32
+ - `croissant_metadata.json` SHA256: `79dca0a5a04e49cde98ddf15228c77d6d3aba5d43568309d0c7919581ae30a17`
33
 
34
  ## Review Use
35
 
croissant_metadata.json CHANGED
@@ -60,7 +60,7 @@
60
  "url": "https://huggingface.co/datasets/SolarSolomon/ra-bench-runtime-authority",
61
  "license": "https://creativecommons.org/licenses/by/4.0/",
62
  "citeAs": "Anonymous. RA Bench Runtime Authority Benchmark for Learned Robot Policies, version 0.1.0, NeurIPS 2026 Evaluations and Datasets supplementary material.",
63
- "citation": "Anonymous submission manuscript: Runtime Authority Bench: An Auditable Evaluation Protocol for Learned Robot Policies in Structured-State Simulation.",
64
  "creator": {
65
  "@type": "Organization",
66
  "name": "Anonymous Authors"
@@ -490,7 +490,7 @@
490
  ],
491
  "rai:uses": "Evaluate runtime authority assignment for learned robot policy proposals in structured-state simulation; reproduce aggregate claims and inspect selected logs.",
492
  "rai:limitations": "The benchmark is simulation-only and does not validate physical robot safety, real perception, ROS2 hardware control, full grasping, lifting, placing, multi-object manipulation, or force-control manipulation.",
493
- "rai:dataCollection": "Episodes and reports are generated from a MacBook-local structured-state MuJoCo tabletop pre-grasp runtime. Selected logs and aggregates are included to keep the review package compact.",
494
  "rai:dataPreprocessing": "Runtime logs are summarized into aggregate CSV/JSON tables, selected representative logs, replay reports, leak-audit reports, safety-invariant reports, and paper figures. Absolute local paths were anonymized.",
495
  "rai:risksAndBiases": "Main risks are overinterpreting simulator evidence as hardware safety, overfitting claims to finite release distributions, and treating replay-only unsafe exposure analyses as executed unsafe robot policies.",
496
  "rai:mitigations": "The package includes explicit scope boundaries, claim-to-evidence mapping, selected-log checks, leak audits, safety-invariant audits, and warnings against real-robot or real-sensor claims.",
 
60
  "url": "https://huggingface.co/datasets/SolarSolomon/ra-bench-runtime-authority",
61
  "license": "https://creativecommons.org/licenses/by/4.0/",
62
  "citeAs": "Anonymous. RA Bench Runtime Authority Benchmark for Learned Robot Policies, version 0.1.0, NeurIPS 2026 Evaluations and Datasets supplementary material.",
63
+ "citation": "Anonymous submission manuscript: Runtime Authority for Learned Robot Policies: An Auditable Evaluation Protocol in Structured-State Simulation.",
64
  "creator": {
65
  "@type": "Organization",
66
  "name": "Anonymous Authors"
 
490
  ],
491
  "rai:uses": "Evaluate runtime authority assignment for learned robot policy proposals in structured-state simulation; reproduce aggregate claims and inspect selected logs.",
492
  "rai:limitations": "The benchmark is simulation-only and does not validate physical robot safety, real perception, ROS2 hardware control, full grasping, lifting, placing, multi-object manipulation, or force-control manipulation.",
493
+ "rai:dataCollection": "Episodes and reports are generated from a local-workstation structured-state MuJoCo tabletop pre-grasp runtime. Selected logs and aggregates are included to keep the review package compact.",
494
  "rai:dataPreprocessing": "Runtime logs are summarized into aggregate CSV/JSON tables, selected representative logs, replay reports, leak-audit reports, safety-invariant reports, and paper figures. Absolute local paths were anonymized.",
495
  "rai:risksAndBiases": "Main risks are overinterpreting simulator evidence as hardware safety, overfitting claims to finite release distributions, and treating replay-only unsafe exposure analyses as executed unsafe robot policies.",
496
  "rai:mitigations": "The package includes explicit scope boundaries, claim-to-evidence mapping, selected-log checks, leak audits, safety-invariant audits, and warnings against real-robot or real-sensor claims.",
supplementary_material.zip CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:37d1c14fa46dc0325e94fdb077d239524d9e3ebdd0e34ea52c93cd9c42719f62
3
- size 26943003
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1610e821b82fa73b2639b4e8200067ee989625153b12115cb99b9d4384c7878b
3
+ size 22733773