PointWorld-BEHAVIOR: 3D Annotations from Robot Simulation Episodes

A packaged release of precomputed 3D annotations derived from BEHAVIOR simulation episodes, used for training and evaluating the PointWorld world model. The dataset, created by NVIDIA, contains episode-level HDF5 files storing robot state, camera parameters, initial RGB-D observations, and rigid-body scene geometry. The repository was last updated on May 7, 2026.

Use Cases

• Training 3D world models based on precomputed robot state and scene geometry annotations.
• Evaluating robotic manipulation policies using derived simulation episode data.
• Benchmarking scene understanding algorithms based on provided RGB-D observations and camera parameters.

Strengths

• Contains precomputed 3D annotations derived from the established BEHAVIOR simulation platform.
• Data is organized as episode-level HDF5 files, which suggests a structured format for sequential tasks.
• Includes multiple data modalities such as robot state, camera parameters, RGB-D observations, and scene geometry.

Limitations

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

Provenance

Source

NVIDIA, derived from the BEHAVIOR simulation platform.

Collection Method

Annotations were precomputed from BEHAVIOR simulation episodes.

Freshness

Last updated 2026-05-07 22:24:15; freshness should be verified.