LeWAM-Simulation-Data
Collection
Simulation training data for LeWAM: OGBench scene and puzzle (3x3). • 10 items • Updated • 1
dataset string | frames int64 | episodes int64 | source dict | files list |
|---|---|---|---|---|
Original LeWM Reacher | 2,010,000 | 10,000 | {
"repo": "quentinll/lewm-reacher",
"revision": "e70a080d0d04c6072123c9ebd343acf7fff28dbf",
"archive_sha256": "4ff2385e49712caa89f21b8e0a246e2614b621d3f22cf2d1224d845e879a1cc2"
} | [
{
"path": "reacher.h5.zst",
"bytes": 22826731465,
"sha256": "68283b9228858115e8e5965cb3025f398f54c57900b48ad279e515d4f64ba20c",
"uncompressed_file": "reacher.h5",
"uncompressed_bytes": 98905882624,
"uncompressed_sha256": "85a7dddfa1801302abcb175a80a23bb69c78291dd977ce40d69aedcb9123da06"
}
... |
LeWAM Reacher
Original LeWM Reacher data, collected with a random policy: 10,000 trajectories and 2,010,000 RGB frames at 224 × 224. Actions have two dimensions. The scene target is hidden during goal-reaching evaluation.
hf download LeWAM/lewam-reacher reacher.h5.zst --repo-type dataset --local-dir "$STABLEWM_HOME/datasets"
zstd -d --rm "$STABLEWM_HOME/datasets/reacher.h5.zst"
manifest.json records source revision, file sizes and SHA-256 hashes. Decompression restores the original reacher.h5 bytes. Episode boundaries are stored in ep_offset and ep_len.
- Downloads last month
- 237