playdata / README.md
gg233gg's picture
Add files using upload-large-folder tool
17098f4 verified
|
Raw History Blame Contribute Delete
2.38 kB
metadata
tags:
  - lerobot
  - robotics
  - agilex

AgileX playdata

LeRobot v3.0 dataset: 62 episodes, 158,682 frames, 30 FPS. Three RGB cameras (front, left, right), 1280×720, H.264 videos. Actions are the original master commands; observation.state contains puppet joint positions. Each has 14 channels: left six joints + left gripper, then right six joints + right gripper. No physical-unit conversion was applied.

Tasks and splits

Original episode numbers start at 1; LeRobot episode_index starts at 0.

Source episodes LeRobot episode_index Task Split
1–30 0–29 play with cup and bowl train
31–34 30–33 play with towel train
35–39 34–38 play with cardbox train
40–42 39–41 play with draw train
43–47 42–46 play with cup and bowl validation
48–52 47–51 play with towel validation
53–57 52–56 play with cardbox validation
58–62 57–61 play with draw validation

The dataset root contains both splits. Select episodes explicitly when loading; loading without an episode selection includes all 62 episodes. Split membership is recorded in meta/info.json and meta/splits.json. The latter also lists the four five-episode validation groups. All task strings retain the requested spelling.

from lerobot.datasets.lerobot_dataset import LeRobotDataset

train = LeRobotDataset("gg233gg/playdata", episodes=list(range(42)))
validation = LeRobotDataset("gg233gg/playdata", episodes=list(range(42, 62)))

meta/stats.json describes the complete dataset. For experiments requiring training-only normalization, derive statistics from the training episodes only.

Provenance and validation

Converted from playdata.tar with the AgileX hdf5_to_lerobot_fast.py converter. Sources were supplied in numeric episode order through temporary zero-padded links, preserving the original HDF5 and images. Task metadata, per-frame task indices and task-index statistics were updated consistently after conversion.

meta/source_episodes.json maps every converted episode to its source, task and split. meta/validation_report.json records frame/episode counts, exact comparison of all converted action/state rows to their float32 source values, video frame counts, and LeRobot decoding checks at the first and last frame of every episode.