LIBERO: Lifelong Robot Learning Benchmark Demonstrations

Provider
Lifelong Robot Learning (UT Austin)
License
CC-BY-4.0 (source)
Size
132 files
Format
HDF5 (in sampled files)
Stored at
Hugging Face (https://huggingface.co/datasets/yifengzhu-hf/LIBERO-datasets)
Access
Free and public. No account with the data provider is needed.
Category
Robotics
Source
https://arxiv.org/abs/2306.03310

About LIBERO: Lifelong Robot Learning Benchmark Demonstrations

The original demonstration datasets for LIBERO, a benchmark for lifelong (continual) robot learning and knowledge transfer in tabletop and kitchen manipulation, built on robosuite. Human teleoperated demonstrations are grouped into the five LIBERO task suites: libero_spatial, libero_object, libero_goal, libero_10 (LIBERO-Long) and libero_90. Each task is one HDF5 file of demonstrations named after its language instruction, for example KITCHEN_SCENE1_open_the_top_drawer_of_the_cabinet_demo.hdf5; the libero_90 folder holds 90 such files. Files contain observation images, proprioception and actions in the robomimic-style HDF5 layout and open with h5py or the LIBERO and robomimic loaders. Measured 2026-11-14 at about 100 GB. LIBERO is the standard simulation benchmark for vision-language-action models such as OpenVLA and pi0, and many LeRobot and RLDS conversions derive from these files. The LIBERO repository licenses the datasets under CC BY 4.0 and the code under MIT; cite Liu et al., LIBERO (NeurIPS 2023 Datasets and Benchmarks).

How to load LIBERO: Lifelong Robot Learning Benchmark Demonstrations in Python

from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="yifengzhu-hf/LIBERO-datasets", filename="libero_10/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it_demo.hdf5", repo_type="dataset")
print(path)

Data structure

File types in sampled files: HDF5

Files

Example file: libero_10/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it_demo.hdf5

Analyze this dataset without downloading it

On Scigantic, LIBERO: Lifelong Robot Learning Benchmark Demonstrations opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.

This page is also available as Markdown and as JSON.

Related Robotics datasets