Implementation:Facebookresearch Habitat lab Datasets download rearrangement
| Knowledge Sources | |
|---|---|
| Domains | Embodied_AI, Data_Management |
| Last Updated | 2026-02-15 02:00 GMT |
Overview
CLI tool invocation for downloading ReplicaCAD/HSSD articulated scene datasets and PDDL task definitions for rearrangement training.
Description
This uses the same datasets_download CLI from habitat-sim but targets rearrangement-specific datasets (ReplicaCAD, HSSD) that include articulated objects with joint definitions. Additionally, PDDL domain and problem YAML files in the habitat-lab repository define the task structure for rearrangement episodes.
Usage
Run before any rearrangement or hierarchical RL training. The ReplicaCAD dataset is the standard benchmark for Habitat rearrangement tasks.
Code Reference
Source Location
- Repository: habitat-sim (download tool)
- File: External CLI +
habitat-lab/habitat/config/habitat/task/rearrange/pddl/set_table.yaml(PDDL definitions)
Signature
python -m habitat_sim.utils.datasets_download \
--uids replica_cad_dataset \
[--data-path <target_directory>]
Import
# CLI tool, no Python import needed
# PDDL files are YAML configs loaded automatically by Habitat
I/O Contract
Inputs
| Name | Type | Required | Description |
|---|---|---|---|
| --uids | str | Yes | replica_cad_dataset or hssd-hab
|
| --data-path | str | No | Target directory (defaults to data/)
|
Outputs
| Name | Type | Description |
|---|---|---|
| Articulated scenes | .glb + .json files | 3D scenes with joint/articulation metadata |
| Object assets | .glb files | Graspable object meshes |
| PDDL definitions | .yaml files | Task domain and problem specifications |
| Episode datasets | .json.gz files | Rearrangement episode definitions |
Usage Examples
Download ReplicaCAD
# Download ReplicaCAD dataset for rearrangement
python -m habitat_sim.utils.datasets_download --uids replica_cad_dataset --data-path data/