Kairos-HomeWorld - A whole-house 3D interactive world model launched by Daxiao Robotics

What is Kairos-HomeWorld?

Kairos-HomeWorld is the world's first fully interactive 3D house model, jointly developed by Daxiao Robotics, the Chinese University of Hong Kong, and Shenzhen Hetao College. The model utilizes a four-stage hierarchical generation architecture to achieve end-to-end generation of a complete 3D residential scene, from text to structure, ensuring physical compliance and enabling interactive objects. It also simultaneously releases open-source datasets of 300,000 real Chinese house floor plans and 5,000 whole-house simulation scenes, providing a localized training foundation for embodied intelligence.

Main functions of Kairos-HomeWorld

• Full-house 3D scene generationIt can generate a coherent, physically reasonable, and fully functional 3D scene of the whole house with one click, covering complex house layouts with multiple rooms, and supporting the full range of sizes from 30㎡ one-bedroom apartments to large flats of over 200㎡.
• Object-level full interactivityEach scene contains an average of more than 15 interactive objects with physical properties (density, hinge structure, material, etc.), which can be directly imported into the simulation engine for operations such as grabbing, moving, stacking, opening and closing.
• Data set specifically for Chinese familiesThe open-source project features 300,000 structured and annotated real-life residential floor plans and 5,000 whole-house simulation scenes, fully recreating the characteristics of traditional Chinese residences such as north-south ventilation, enclosed kitchens, and independent balconies.
• Robot simulation trainingIt supports full-process simulation training for robots to complete complex and long-term household chores such as cross-room navigation and multi-room item organization, significantly shortening the migration cycle from virtual to reality.

The technical principles of Kairos-HomeWorld

• Four-stage layered generation architectureThe framework decouples the complex task of generating a whole house into four progressive stages: The first stage, based on the K-D tree structured representation method, transforms the real residential floor plan into a hierarchical text structure that can be efficiently learned by a large language model, avoiding room overlap and topological breaks; The second stage adopts a hierarchical strategy of "top-view global initialization + first-person detailed walkthrough" to anchor the generation process with the 3D building shell, solving the geometric drift problem in the 2D to 3D upgrade; The third stage constructs a recursive closed-loop verification mechanism by fine-tuning the visual language model to automatically detect and correct physical violations such as "sofa blocking doors" and "objects passing through walls"; The fourth stage generates interactive objects with complete physical properties for the desktop and tabletop through a surface center object placement algorithm.
• Physical compliance and closed-loop verification mechanismThe system uses a finely tuned VLM to perform visual analysis on the generated scene, identify layout conflicts and physical errors, and propose corrective actions through reflection loops to control the furniture layout collision rate at the industry's best level, ensuring that the scene conforms to real physical rules.
• Interactive object asset generationWith the help of the Physx-Omni model, the system automatically assigns physical properties such as material, density, hinge, and manifold to objects, generating an average of more than 15 operable objects. All assets can be directly instantiated into the simulation engine, supporting realistic physical simulations such as collision, gravity, friction, and fluid.

How to use Kairos-HomeWorld

• Resource AcquisitionVisit the Kairos-HomeWorld project homepage https://kairos-homeworld.github.io/ to download the open-source dataset, model weights, and technical documentation.
• Environment configurationDeploy a simulation platform that supports Physx-Omni and configure GPU computing resources.
• Scene generationInputting natural language commands triggers a four-stage layered generation process, automatically completing the entire chain of building from the apartment layout to an interactive 3D scene.
• Simulation DeploymentImport the generated complete scene into the simulation engine and instantiate interactive object assets with physical properties.
• Training execution: Connect to the robot model and issue household chores instructions, and perform cross-room navigation and object operation training in a virtual environment.
• Relocation and LandingThe training strategy was extracted, transferred to real robot hardware, and validated and fine-tuned in a real home environment.

Kairos-HomeWorld's core advantages

• Localized data foundation: The world's first dataset specifically designed for Chinese families covers local characteristics such as north-south ventilation, wet and dry separation, and non-rectangular kitchens, solving the problem of European and American datasets not being suitable for Chinese families.
• Scale and cost leadership: The dataset of 300,000 real floor plans is the largest of its kind in the world. Virtual generation has almost zero marginal cost, is not limited by the total number of real housing units, and its scalability far exceeds that of on-site collection.
• Global consistency breakthrough: It pioneered a unified generation framework for the entire house, solving the industry bottleneck of traditional methods that can only cover a single room and lack global consistency and operability.
• End-to-end interactive: Breaking the limitation of only being able to see but not use, it achieves full-link automation from text to trainable scenarios, significantly reducing the threshold for embodied intelligence R&D.

Kairos-HomeWorld project address

• Project official websitehttps://kairos-homeworld.github.io/
• GitHub repositoryhttps://github.com/Kairos-HomeWorld/HomeWorld
• arXiv technical paper: https://arxiv.org/pdf/2606.06390

Kairos-HomeWorld's Competitive Product Comparison

Application Scenarios of Kairos-HomeWorld

• Training for home service robots: It provides a high-fidelity simulated Chinese home environment for sweeping, tidying, and companion robots, and supports long-distance task training across rooms.
• Embodied Intelligence R&D Platform: As a virtual training ground, it accelerates the iteration of embodied AI algorithms and reduces the migration cost and R&D threshold from simulation to real machines.
• Smart Home Pre-Demonstration Verification: Simulate furniture layout and equipment movement in a virtual environment to optimize interior space planning and smart home solution design.
• Academic research and benchmarking: It provides large-scale standardized datasets and evaluation benchmarks for research on indoor scene understanding, robot path planning, and object manipulation.
• Real Estate & Home Improvement Preview: Quickly generate interactive 3D house viewing models based on real house layouts, supporting decoration plan previews and space function verification.