Karios · HomeWorld | Whole-Home Scene Generation

Large-scale Datasets

300K

Real-World Floorplan Dataset

Chinese Style
Richly Captioned
Fully Vectorized

Furnished Whole-Home Dataset

Complex Geometry
Manipulable Objects
Full 3D & Sim-ready

Whole-home Scene Generation

Loading…

Embodied Interaction Demo in Generated Scenes

Highlights

Unified whole-home generation pipeline — From a text prompt, generates simulation-ready, fully furnished whole-home scenes via four controllable stages: floorplan generation → furniture layout → refinement → manipulable object placement.
Embodied-AI ready — Surface-centric small-object placement, basic physical attributes, textures for downstream simulation.
Large-scale whole-home scenes — 300K real, vectorized residential floorplans and 5K furnished whole-home 3D scenes with manipulable objects.
Controllable floorplans — LLM trained with K-D tree representation and rich captions for fine-grained, prompt-conditioned layout generation.
Image-driven hierarchical furnishing and object placement — Foundation image models plus top-down and ego-centric view roaming, grounded in an explicit 3D shell for cross-view consistency.
VLM recursive refinement — Iteratively fixes collisions, blocked doorways, and layout violations for physically plausible scenes.

Read full abstract

Indoor scene generation is crucial for robot simulation and modern interior design. However, complex layouts together with scarce 3D scene data make learning-based generation challenging. Existing methods often rely on hand-crafted rules or focus on isolated sub-tasks (e.g., floorplan synthesis or single-room furnishing), producing whole-home scenes that lack global coherence, realism, and simulation readiness. To mitigate these limitations, we propose a unified hierarchical framework that decomposes indoor scene synthesis into controllable stages. First, we curate a large-scale dataset of 300K real residential floorplans to train a large language model for whole-home floorplan generation. With detailed descriptions and a K-D tree–based representation, our method enables fine-grained, controllable whole-home floorplan generation. Building upon the generated whole-home floorplan, we leverage image generation models to draft furniture layouts from multi-level roaming viewpoints, and then generate the layouts of small manipulable objects on different supporting surfaces (e.g. cabinets, desks, and dining tables) for embodied AI simulation. During furniture & object layout generation, a VLM-based refiner iteratively corrects furniture & object placement, and a 3D generative model enables flexible replacement of individual assets. We further attach basic physical attributes and simple surface texture and lighting setups to complete the pipeline for embodied AI use. Experiments and user studies demonstrate that our pipeline produces indoor spaces with greater layout diversity and stronger 3D design appeal, outperforming prior methods on both quantitative and qualitative metrics. Finally, alongside our generation pipeline, we will release the floorplan dataset and 5K fully furnished scenes to the community.

Method

Four-stage whole-home indoor scene generation pipeline — Figure 1 — Overview of the whole-home indoor scene generation pipeline.

Stage 1: Floorplan Generation — Curate 300K real residential floorplans and train an LLM with K-D tree representation for prompt-conditioned, controllable whole-home layouts.
Stage 2: Image-Driven Hierarchical Furnishing — Instantiate an unfurnished 3D shell, then place furniture via hierarchical view roaming (top-down → ego-centric) with grounding and asset retrieval.
Stage 3: Recursive Refinement — A finetuned VLM refiner detects collisions, blocked doorways, and violations, then iteratively proposes corrective actions in a reflective loop.
Stage 4: Manipulable Object Placement — Surface-centric synthesis populates desks, counters, and shelves with small objects, with physical attributes for simulation-ready embodied AI.

Generation Results

Whole-home Samples

Functional Room Samples

Generated functional rooms: living room, bedroom, bathroom, and kitchen — For each room, the view inside the red dashed box below is a zoom-in focused on manipulable objects.

BibTeX

@misc{li2026homeworld,
  title         = {HomeWorld: A Unified Floorplan-to-Furnished Framework for Generating Controllable, Densely Interactive Whole-Home Scenes},
  author        = {Wenbo Li and Xiaoliang Ju and Zipeng Qin and Rongyao Fang and Hongsheng Li},
  year          = {2026},
  eprint        = {2606.06390},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2606.06390},
  note          = {Project page: https://kairos-homeworld.github.io/}
}