Chinese researchers unveil AI framework for generating whole-home robot training scenes

A team of Chinese researchers has developed Kairos-HomeWorld, the first unified framework capable of generating coherent, simulation-ready home environments from simple text prompts. The framework addresses a long-standing data bottleneck in training household robots by producing whole-home-scale residential scenes with multiple manipulable objects, moving beyond conventional single-room layouts. The research involves Ace Robotics (backed by SenseTime), the Multimedia Laboratory at Chinese University of Hong Kong, and Shenzhen Loop Area Institute. The four-stage process includes floor plan construction, 2D-to-3D conversion, furniture layout generation, refinement, and final object-level generation, with each scene averaging over 15 manipulable objects. Ace Robotics announced the breakthrough on Friday, stating that these high-fidelity simulations provide a robust foundation for advancing embodied intelligence and accelerating real-world robot training.