Breakthrough in Training Household Robots Using
AI-Generated Homes, Chinese Researchers Say
A team of Chinese researchers has announced a breakthrough in training robots within real-world home environments, potentially overcoming a long-standing data bottleneck and accelerating the adoption of domestic robots. The innovation comes from Ace Robotics, a start-up backed by Hong Kong-listed artificial intelligence company SenseTime, in collaboration with the Multimedia Laboratory at the Chinese University of Hong Kong and the Shenzhen Loop Area Institute.
According to the South China Morning Post, the researchers introduced “Kairos-HomeWorld,” which they describe as the world’s first unified framework capable of generating coherent, accurate, and simulation-ready home environments using simple text prompts. The framework is specifically designed to break the constraints of conventional indoor scene generation, which has traditionally been confined to single-room layouts and limited interactivity.
Rather than isolating individual rooms, Kairos-HomeWorld generates entire home-scale and object-level residential scenes suitable for training both domestic robots and humanoids. The platform relies on a four-stage process: starting with floor plan construction, progressing through 2D-to-3D conversion and furniture layout generation, and moving through a refinement stage before final object-level generation. On average, each generated scene incorporates more than 15 manipulable objects.
“These high-fidelity, large-scale simulations provide a robust foundation for advancing embodied intelligence applications and accelerating real-world robot training,” Ace Robotics stated in its announcement.