retrieval-try-1-raw

Raw recorded trajectories of MikasaCabinetRetrieval-v0 — 1000 episodes, seeds 1100–2109 with gaps, 10 Hz, every one a success. About 30 GiB.

For a training-ready version of the same 1000 episodes in LeRobot format (parquet + h264 video, 2.6 GB instead of 30), see nurtayev-d/retrieval-try-1.

A second task recorded the same way, MikasaSeasonDish-v0, is in nurtayev-d/seasondish-try-1.

Layout

seed_<N>/<timestamp>.h5      actions, agent state, both camera streams, full scene states
seed_<N>/<timestamp>.json    episode metadata: env config, seed, elapsed steps, success, provenance
seed_<N>.log                 per-episode recorder output (at the root, not inside the episode folder)
pool.env                     provenance of the whole 1000-episode set — read this first
_pools/<pool>/               each source pool's own job list and conversion logs (see below)

Provenance

Two robots are involved, and that is deliberate. Planning ran on ds_fetch with state observations only; the frames were then produced by replaying each planned trajectory on fetch_cam224 — ManiSkill's own stock Fetch with 224×224 cameras, nothing of ours in the robot itself. Full commits and timestamps for both passes are in pool.env.

The episodes come from two recording pools, joined here into one set: 991 from retrieval-stock224 and 9 from retrieval-stock224-topup0 (seeds 2100, 2101 and 2103–2109), the top-up existing only to bring the set to exactly 1000 kept episodes. Both pools ship a jobs.txt, rerender.env, done and — for the main pool — a convert.log under the same names, so they are kept apart under _pools/<pool name>/ rather than overwriting each other at the root. The root pool.env describes the union and is the one to read.

Notes

• 97.6% of each .h5 is the two RGB streams, already gzip-compressed inside HDF5; every other array together is about 2%. If you only need scene states or actions, this repo is an expensive way to get them — and the frames are far smaller in the LeRobot release.
• Scene states — poses of the cup and of every kitchen articulation — are here and not in the LeRobot version. That is the main reason to prefer this repo.
• Each episode's .json names the exact planning recording it was replayed from.