On September 15, Guangzhou Shangpin Home Collection Co.,Ltd. (referred to as Shangpin Home Collection) jointly launched the WorldSimReady-Home simulation dataset for embodied intelligence in home services with Yootta. The first phase of the dataset is now available on the WorldSimReady-Home platform, Hugging Face, and ModelScope, offered free of charge to robot hardware teams, embodied intelligence and algorithm companies, research institutions, and developers at large. It provides a large-scale, high-quality data foundation for simulation training, algorithm evaluation, and policy iteration of home service robots.
This marks another substantive step by Shangpin Home Collection in the embodied intelligence field. From strategic cooperation with Qiyuan Robot and the launch of the world's first “Robot-Ready Home” physical store experience, to the open-sourcing of the home simulation dataset, Shangpin Home Collection is leveraging its “scenario + data” capabilities to bridge the entire journey of robots from laboratory to household environments.
Building China's largest high-quality simulation dataset for home embodied intelligence The biggest obstacle to humanoid robots entering households is not hardware, but the complexity of home environments. Layouts differ, furniture arrangements vary, and daily movement patterns are intricate. Robots may excel in laboratory settings but often stumble in real homes, a challenge the industry calls the “Sim2Real transfer gap”. To close this gap, robots need repeated training in environments close to real homes, but real-machine data collection is costly and scenario replication is difficult. Consequently, high-quality home simulation data has become critical infrastructure for the industry.
Shangpin Home Collection's exploration in embodied intelligence has been underway for some time. In July of this year, the company entered a strategic partnership with Shanghai Wei New Materials Qiyuan Robot to co-develop the “Robot-Ready Home” concept. On September 5, the world's first “Robot-Ready Home” physical experience store began trial operations at Grandview Mall in Guangzhou, placing humanoid robots in a real home setting for public interaction. Furthermore, Shangpin Home Collection has been advancing its second track in physical AI robot data, drawing on over 3 million real floor plans and more than 30 million design blueprints to create a digital scenario foundation that robots can recognize and use for training.
“Robot-Ready Home” addresses how robots can physically fit into and navigate a space, while WorldSimReady-Home tackles how robots can train extensively and learn accurately in digital environments before stepping into a real home. The former provides the spatial foundation, and the latter the data foundation, together completing Shangpin Home Collection's capability framework in embodied intelligence. The joint open-sourcing initiative with Yootta marks a key transition of this data capability from internal accumulation to industry-wide sharing.
Yootta is a 4D simulation data infrastructure provider for physical AI, offering a differentiated 4D ground-truth technology system and its self-developed Yootta full-chain tool suite, covering products from real-scene reconstruction, 4D SimReady data, world model training, to closed-loop simulation verification. The collaboration leverages complementary strengths: Shangpin Home Collection contributes its years of accumulated real home spaces and high-quality 3D design assets, while Yootta brings full-chain technical capabilities from real-scene reconstruction and simulation training to robot policy deployment. Together, they aim to build China's largest high-quality simulation dataset, WorldSimReady-Home, centered on “scenario + assets + tasks”.
A 100,000-square-meter high-fidelity home scenario as a robot training ground The first phase of the WorldSimReady-Home open-source dataset includes 100,000 square meters of high-fidelity home scenarios, 10,000 interactive assets, and 1,000 standardized robot simulation task examples. This is not a static library. Using the initially open-sourced home scenarios as a “seed”, and leveraging multi-dimensional variations in spatial layout, object placement, materials, and task conditions, the dataset can expand to generate tens of thousands of differentiated home simulation environments tailored to robots of various forms and functions.
The over 3 million real floor plans and 30 million design blueprints supplied by Shangpin Home Collection provide a continuous source of real home spaces and high-quality 3D design assets. In terms of data quality, WorldSimReady-Home is built on real home design data, with full-dimensional structuring of geometry, semantics, physics, and tasks. Scene geometry reaches millimeter-level precision, semantic and material accuracy exceeds 99%, and key physical property accuracy exceeds 95%. Each asset carries four core layers of information: geometric accuracy, semantic completeness, physical plausibility, and interaction readiness, including multi-modal data such as RGB images, depth maps, and semantic segmentation, as well as physical and interactive properties like object joints, mass, friction, damping, and collisions.
In simple terms, this equates to building a “digital home training ground” for robots. Here, robots can repeatedly practice tasks such as navigation, grasping, placement, and long-horizon household chores without risking damage to furniture or accidents. Once training is complete, the learned policies can be transferred to physical robots, enabling a “simulate first, then deploy” approach.
Cross-embodiment, cross-scenario, and cross-task for versatile application Data collected from real machines traditionally adapts only to specific robot platforms. WorldSimReady-Home has engineered a decoupling of environments from robot bodies at the engineering level, supporting quick integration of standard models like URDF. Humanoid robots, quadruped robots, and wheeled robots can all reuse the same set of scenario and task data. This means a single suite of home scenario data can simultaneously serve training needs across different robot forms and functions.
Developers can directly use the dataset to conduct simulation training for navigation, object manipulation, and complex long-horizon household tasks. Compared to debugging on physical robots, simulation training offers lower trial-and-error costs, controllable experimental conditions, and large-scale task reproducibility, allowing extensive training and evaluation to be completed in virtual environments.
The home scenario is the starting point of the SimReady open-sourcing initiative. As more hardware, model, and algorithm teams, along with scenario providers, join, datasets for industrial, commercial, and special-use scenarios will roll out progressively, strengthening the public data foundation for Physical AI.
Open-sourcing as co-creation to drive embodied intelligence into households This open-source move extends Shangpin Home Collection's “scenario-data-ecosystem” pathway: on the scenario side, its three decades of home furnishing expertise create the spatial foundation for robots entering residences; on the data side, real floor plans and design blueprints are converted into trainable, generalizable simulation assets, reinforcing the “spatial intelligence digital foundation” and offering scarce home training resources to the industry; and on the ecosystem side, open-sourcing drives collaborative industry building.
From real homes to simulated worlds, and back again: Shangpin Home Collection is paving the most challenging path to bring robots into everyday households. “Robot-Ready Home” readies the home for robots, and “WorldSimReady-Home” readies the robots before they arrive. One in the physical realm and one in the digital realm, both aim for the same goal: to make robots a genuine part of family life. Looking ahead, Shangpin Home Collection will continue advancing the development and openness of its embodied data system, expanding additional scenario datasets, and leveraging dual innovation in scenarios and data to accelerate the arrival of a human-robot cohabitation lifestyle.