format_version int64 | artifact string | arms dict | evaluation dict | content dict | scorer dict | self_test dict |
|---|---|---|---|---|---|---|
1 | WARP-RM paper-A simulation traces | {
"vanilla": {
"directory_glob": "fullhz_vanilla_sh*",
"worlds": 128
},
"warp_rm": {
"directory_glob": "fullhz_warp_sh*",
"worlds": 128
}
} | {
"task": "sim_put_the_plastic_bottles_in_the_bin",
"seed_first": 20260511,
"seed_last": 20260638,
"horizon_seconds": 60,
"fps": 30,
"bottle_count": 6,
"trace_format": "compressed NumPy qpos arrays"
} | {
"qpos_trace_count": 256,
"tree_sha256": "c2963d58fdbfb3debf3c59503ae361bde08c7c3e452ec1cadd1bcb61e34b8eb7",
"tree_hash_method": "sorted qpos trace paths, each file's sha256sum line, then sha256"
} | {
"repository": "https://github.com/uynitsuj/abc-rabc",
"commit": "eab8e9b3bdfa568cd749a56fdc63f151c3a9f166",
"script": "score_bottles.py",
"rule": "tight-anypart"
} | {
"passed": true,
"baseline_bottles": 509,
"warp_rm_bottles": 598,
"baseline_bottles_per_episode": 3.977,
"warp_rm_bottles_per_episode": 4.672,
"paired_difference": 0.695,
"paired_t": 4.471,
"baseline_time_per_bottle_seconds": 10.688,
"warp_rm_time_per_bottle_seconds": 8.751,
"baseline_throughput_bo... |
WARP-RM paper simulation n=128 traces
Canonical paper-A rollout traces for the two released simulation policies over the fixed 128-episode evaluation suite. They are the original trace pair used by the public evaluator self-test, rather than a later re-run.
The archive contains 256 qpos traces: 128 vanilla-policy traces and 128 WARP-RM-policy traces. It is deliberately limited to evaluation output (no model weights, DINO features, or raw source-capture data).
With the released score_bottles.py --self-test, these traces reproduce the
paper table exactly:
| metric | vanilla | WARP-RM |
|---|---|---|
| bottles retrieved (n=128) | 509 (3.977/episode) | 598 (4.672/episode) |
| time per bottle (s) | 10.688 | 8.751 |
| throughput (bottles/hour) | 243.022 | 301.282 |
| all six bottles (%) | 12.500 | 28.125 |
MANIFEST.json records the content hash, scorer revision, evaluation
configuration, and the associated public artifact revisions.
The companion release code recomputes these metrics from the traces without private data or services.
- Downloads last month
- 338