THULab/so100_test

SO100 Test (TsFile)

Apache TsFile version of burgalon/so100_test, a LeRobot v2.1 SO100 tutorial trajectory dataset. Each frame contains a six-dimensional robot action and a six-dimensional observed robot state.

Overview

• Train split: 53 episodes and 23,142 frames (0:53).
• Sampling rate: 30 fps.
• Tasks: one task.
• Robot type: SO100.
• Converted layout: one data/so100_test_train.tsfile containing all train episodes as separate TAG devices.

Schema (TsFile structure)

• Time (INT64, milliseconds) is round(timestamp * 1000) and restarts for each episode.
• episode_index and task_index (TAG) identify the source episode and task. A single episode can be selected with WHERE episode_index='0' AND task_index='0'.
• frame_index and sample_index (FIELD, INT64) preserve source frame and global sample ordering; source index is renamed to sample_index.
• observation_state_0..5 and action_0..5 (FIELD, FLOAT) are the flattened six-dimensional source vectors.

The source timestamp column is omitted after conversion to Time, as approved for this conversion. The 53 source Parquet files are merged without dropping rows. Camera videos are not uploaded; they remain in the original videos/ tree.

Usage

Read data/so100_test_train.tsfile with the Apache TsFile Java or Python SDK.

Source & license

• Original dataset: https://huggingface.co/datasets/burgalon/so100_test
• Author / publisher: burgalon
• Paper: not declared by the original dataset
• License: Apache-2.0