InteriorAgent: Interactive USD Interior Scenes for Isaac Sim-based Simulation

InteriorAgent is a collection of high-quality 3D USD assets specifically designed for indoor simulation in NVIDIA Isaac Sim environments. Each asset is structured with modular materials, scene description files, and physics-ready geometry, enabling fast integration for embodied AI and robotics tasks such as navigation, manipulation, and layout understanding.

A sample scene from the InteriorAgent dataset rendered in Isaac Sim. The scene features high-quality 3D assets such as sofas, cushions, tables, and chandeliers, all modeled with real-world scale. The bottom panel shows loaded asset files (e.g., kuijiale_0021.usda), and the right panel displays a hierarchical list of all 3D objects along with their semantic labels, supporting spatial reasoning and interaction in embodied AI tasks.

🚀 Features

• ✅ Fully compatible with Isaac Sim 4.2 and 4.5 on both Windows and Linux.
• 🎮 Built for real-time simulation, supports interactive physical agents.
• 🧱 Material system based on NVIDIA MDL (Material Definition Language), ensures photorealistic rendering and cross-version compatibility.
• 📦 Provided in .usd and .usda format with structured folders for materials, meshes, lighting, and floorplan.

---

🗂 Directory Structure

The dataset is organized per scene. Each scene folder follows the structure below:

kujiale_xxxx/
├── .thumbs/ # Optional thumbnail or cache folder (can be ignored)
├── Materials/ # Material library
│ ├── Textures/ # Texture images (optional, omitted here)
│ ├── *.mdl # MDL material and instance files
├── Meshes/ # Mesh geometry (e.g., .usd or .obj)
├── kujiale_xxxx.usda # Top-level USD scene file
├── limpopo_golf_course_4k.hdr # Environment lighting HDR file
└── rooms.json # Room-level metadata and spatial layout (JSON format)

🧭 Room Metadata (rooms.json)

Each scene folder includes a rooms.json file that defines the 2D floorplan layout of the space. It contains a list of room entries, where each room is defined by:

room_type: the semantic label (e.g., "living_room", "bedroom", "balcony", etc.)

polygon: a list of 2D coordinates representing the room's floor boundary in world coordinates

📌 Example

{
    "room_type": "balcony",
    "polygon": [
        [-0.3784970703125, -6.55287060546875],
        [4.005734375,      -6.55287060546875],
        [4.005734375,      -4.8603486328125],
        [-0.3784970703125, -4.8603486328125]
    ]
}

This represents a balcony room with a rectangular floorplan defined by a clockwise polygon in the Isaac Sim world coordinate system (X-Y plane). The polygon can be visualized or parsed using any geometric library (e.g., Shapely) to determine area, intersection, adjacency, etc.

🧪 Integration Tips

The coordinate system is consistent with Isaac Sim’s world frame: X is forward, Y is right, Z is upward.

Room geometry can be directly loaded using libraries like Shapely for spatial reasoning or map generation.

📦 Usage in Python

from shapely.geometry import Polygon
import json

with open("rooms.json", "r") as f:
    rooms = json.load(f)

for room in rooms:
    poly = Polygon(room["polygon"])
    print(f"Room: {room['room_type']}, Area: {poly.area}")

A hierarchical view of structural elements in an InteriorAgent scene. All architectural components are grouped under four main semantic categories: ceiling, wall, floor, and other (including door and window).

🛠 Compatibility

• ✅ Tested with:
  • Isaac Sim v4.2
  • Isaac Sim v4.5
  • Operating Systems: Windows 10/11, Ubuntu 22.04
• 🔧 MDL materials tested with Omniverse RTX renderer.
• 🌐 All files are offline usable and require no additional dependencies.

🏠 Citation

If you use InteriorAgent in your research or development, please cite or link to our project page:

@misc{InteriorAgent2025,
  title        = {InteriorAgent: Interactive USD Interior Scenes for Isaac Sim-based Simulation},
  author       = {SpatialVerse Research Team, Manycore Tech Inc.},
  year         = {2025},
  howpublished = {\url{https://huggingface.co/datasets/spatialverse/InteriorAgent}}
}

📄 License

This dataset is released under InteriorAgent License.