Gasgoo Munich-Lightwheel Intelligence has struck a strategic partnership with Shengshu AI, Gasgoo has learned. The duo will collaborate deeply on establishing data standards for general world models, building robust evaluation systems, and driving real-world implementation. Together, they aim to accelerate the path toward models that are "physically credible, task-capable, and verifiable in real-world scenarios."
Image Credit: Lightwheel Intelligence
Specifically, to meet the demands of training and iterating general world models, the partners will combine Lightwheel’s expertise in defining human video and synthetic simulation data with Shengshu’s technical prowess in absorbing multi-source heterogeneous data. Together, they plan to explore data standards and construction methodologies, aiming to foster efficient data coordination. The focus is on enhancing the model's ability to interpret physical space-time, causal logic, and dynamic evolutionary processes.
Regarding evaluation systems, the companies will leverage Lightwheel’s industrial-grade simulation platform, RoboFinals. They aim to define evaluation tasks, metrics, and data quality standards for world and embodied intelligence models, creating a continuous closed-loop cycle that spans capability diagnosis, data supplementation, and model optimization.
To drive continuous iteration, the two will explore applications in embodied intelligence and physical AI. Their collaboration will cover the full spectrum: data collection, standard operating procedure (SOP) conversion, simulation testing, real-machine verification, and feedback assessment. Lightwheel will contribute simulation scenarios, evaluation platforms, and verification capabilities, while Shengshu will bring its general world model's strengths in spatial understanding, physical law modeling, and action outcome prediction.
Lightwheel Intelligence focuses on building data and evaluation infrastructure for physical AI. Built on its proprietary "Solve-Measure-Generate" full-stack simulation platform, the company provides high-quality, large-scale synthetic data, industrial-grade simulation evaluation, human video data, and deployment feedback capabilities.
Shengshu AI is dedicated to constructing a general intelligence framework that models, reasons, predicts, and acts upon the laws of both digital and physical worlds. Centered around its world action model, MotuBrain, the company is developing an embodied intelligent general brain that unifies "world prediction" and "action execution."
According to Xie Chen, founder and CEO of Lightwheel Intelligence, data is set to become the core infrastructure determining large-scale deployment as the industry enters the physical AI stage.
Unlike large language models, which leverage the internet as natural training material, or autonomous driving, which relies on real-world data loops from production fleets and drivers, physical AI deals with more than just text, images, or visual predictions. It involves continuous physical interaction between hands and objects, or robots and environments—complex processes like grasping, pushing, assembly, deformation, friction, and collision. Given this distinction, physical AI requires at least 1 billion data generators capable of continuous production.
