Datasets:

superbigsaw
/

MansionWorld

license: cc-by-4.0
task_categories:
  - robotics
tags:
  - embodied-ai
  - simulation
  - 3d-scene
  - ai2thor
  - multi-floor
  - indoor-environment
pretty_name: MansionWorld
size_categories:
  - 1K<n<10K
extra_gated_prompt: >-
  By accessing this dataset, you agree to use it for research purposes only and
  to cite the associated paper in any publications.
extra_gated_fields:
  Name: text
  Affiliation: text
  Purpose of use: text
  I agree to use this dataset for non-commercial research purposes only: checkbox

MansionWorld

MansionWorld is a large-scale multi-floor indoor scene dataset for embodied AI research, built on top of the AI2-THOR simulator.

Key Features

1,000 interactive multi-floor buildings with 2–10 floors each
10,000+ rooms in total across all buildings
Diverse building types: Residential (50%), Office (30%), Public (20%)

Download

The dataset is provided as zip archives for easier downloading:

File	Size	Description
`mansionworld.zip`	~2.9 GB	All 1,000 building scenes
`mansion_patch.zip`	~789 MB	Additional assets (elevators, stairs, etc.)

Dataset Structure

After extracting mansionworld.zip:

mansionworld/
├── {building_type}_{floors}f_{area}_fp{id}#{variant}/
│   ├── floor_1.json      # Floor configuration (doors, walls, objects, rooms)
│   ├── floor_1.png       # Floor plan visualization
│   ├── floor_2.json
│   ├── floor_2.png
│   └── ...
└── ...

Building Naming Convention

Example: office_corporate_hq_6f_300_fp001#1

office_corporate_hq: Building type (office - corporate headquarters)
6f: Number of floors (6 floors)
300: Approximate floor area (m²)
fp001: Floorplan template ID
#1: Variant index

Usage

Coming soon: Code examples for loading MansionWorld scenes in AI2-THOR.

License

This dataset is released under the CC BY 4.0 License.