THULab/libero_10

LIBERO-10 / LIBERO-Long (TsFile)

Apache TsFile version of lerobot/libero_10.

Overview

LIBERO-10 (the LIBERO-Long suite) is the long-horizon split of the LIBERO lifelong-robot-learning benchmark. It contains human teleoperation demonstrations of 10 distinct long-horizon manipulation tasks performed on a simulated Franka Emika Panda arm — each task chains several sub-goals (e.g. "put both moka pots on the stove", "put the black bowl in the bottom drawer of the cabinet and close it").

• Tasks: 10 long-horizon manipulation tasks (task_index 0–9).
• Episodes: 379 demonstrations.
• Frames: 101,469 timesteps total, recorded at 10 fps.
• Robot: Franka Emika Panda (robot_type = panda).
• Source format: LeRobot v3.0.

The 10 tasks (task_index → instruction):

task_index	instruction
0	put the white mug on the left plate and put the yellow and white mug on the right plate
1	put the white mug on the plate and put the chocolate pudding to the right of the plate
2	put the yellow and white mug in the microwave and close it
3	turn on the stove and put the moka pot on it
4	put both the alphabet soup and the cream cheese box in the basket
5	put both the alphabet soup and the tomato sauce in the basket
6	put both moka pots on the stove
7	put both the cream cheese box and the butter in the basket
8	put the black bowl in the bottom drawer of the cabinet and close it
9	pick up the book and place it in the back compartment of the caddy

Schema (TsFile structure)

All 379 episodes are stored in the TsFile table model, using episode_index and task_index as TAG (device) dimensions.

• Time (INT64, milliseconds) — per-frame timestamp, restarts at 0 for each episode (round(timestamp * 1000); 100 ms apart at 10 fps).
• episode_index (TAG) — demonstration id (0–378). Query a single episode with WHERE episode_index=0.
• task_index (TAG) — task id (0–9). Query all demos of one task with WHERE task_index=3.
• frame_index (FIELD, INT64) — frame number within the episode.
• sample_index (FIELD, INT64) — the source global index column.
• observation_state_0 .. observation_state_7 (FIELD, FLOAT) — 8-D proprioceptive robot state.
• action_0 .. action_6 (FIELD, FLOAT) — 7-DOF action.

The source timestamp column is dropped because it equals Time / 1000. The converted table is stored across multiple .tsfile shards (data/libero_10_*.tsfile); together they hold all 101,469 rows.

Usage

Read the .tsfile files with the Apache TsFile Java or Python SDK.

# pip install tsfile
from tsfile import TsFileReader
reader = TsFileReader("data/libero_10_1.tsfile")
# query one episode: WHERE episode_index=0

Videos

The original dataset records two camera streams per episode (observation.images.image, observation.images.wrist_image). These MP4 videos are not included here — they are not time-series data. Retrieve them from the original dataset under lerobot/libero_10/videos.

Source & license

• Original dataset: https://huggingface.co/datasets/lerobot/libero_10
• Author / publisher: lerobot (LeRobot team)
• Benchmark: LIBERO — Benchmarking Knowledge Transfer for Lifelong Robot Learning, Liu et al., 2023 (arXiv:2306.03310).
• License: Apache-2.0 (inherited from the original dataset)