LIBERO-3D: Processed Robot Demonstration Data for 3D-Aware Learning

LIBERO-3D is a processed robot demonstration dataset packaged for 3D-aware robot learning. It contains trajectories from the original LIBERO task suites, converted into per-episode files with multiple sensor modalities. The dataset was authored by Ayana666888 and last updated on Hugging Face in May 2026.

Use Cases

• Train robot manipulation policies based on multi-modal demonstration trajectories.
• Develop 3D scene understanding models based on RGB, depth, and segmentation data.
• Benchmark robot learning algorithms based on the standardized LIBERO task suites.
• Simulate robot actions based on recorded state and action sequences.

Strengths

• Contains multiple aligned sensor modalities per episode, including RGB, depth, and segmentation.
• Built upon the established LIBERO task suites, providing a known benchmark structure.
• Data is packaged into per-episode .npz files for convenient loading.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Processed from the original LIBERO demonstration dataset.

Collection Method

Trajectories were converted into per-episode .npz files.

Freshness

Last updated 2026-05-14 16:34:07; freshness should be verified.