LIBERO: Robot Manipulation Demonstrations for Lifelong Learning

LIBERO contains robot manipulation demonstrations organized into five task suites for lifelong learning research, including the LIBERO-90 and LIBERO-spatial sets. Created by Yifeng Zhu and updated in May 2025, the data provides expert trajectories stored in HDF5 format for benchmarking imitation learning algorithms.

Use Cases

• Training imitation learning policies using the HDF5 demonstration trajectories
• Benchmarking lifelong learning algorithms across the LIBERO-90 sequential task suite
• Evaluating spatial generalization in robotics using the libero_spatial task demonstrations

Strengths

• Organized into 5 distinct task suites for modular benchmarking
• Includes LIBERO-90 partition containing 90 unique manipulation tasks
• Released under Apache 2.0 license for open research use

Limitations

• Metadata regarding specific sensor inputs or action space dimensions is not provided in the repository description
• Requires external scripts from the official LIBERO repository for proper data loading and environment synchronization

Provenance

Source

Yifeng Zhu, LIBERO Project

Collection Method

Expert demonstrations

Freshness

Last updated May 2025.