At NVIDIA GTC 2026 in San Jose, XGRIDS presented its Real2Sim workflow, designed to capture physical environments and convert them into digital world models that can be used for robot simulation, training and validation. In this process, data from sensors such as lidar and cameras is used to reconstruct real spaces as 3D environments, allowing developers to build simulation scenes based on real-world conditions rather than manual 3D modeling.
the announcement is part of XGRIDs spatial intelligence offering for robotics and physical AI. The company said its platform supports NVIDIA Omniverse NuRec for OpenUSD-based rendering and was shown during a startup pitch, robotics demonstrations within NVIDIA’s ecosystem, and a joint showcase with Amazon Web Services.
According to XGRIDS, its system combines lidar, computer vision and 3D reconstruction to transform physical spaces into simulation-ready environments. The company said this can reduce the work involved in building high-fidelity digital scenes and make it easier to update those scenes when the real environment changes.
During its GTC demonstrations, XGRIDS also showed its spatial perception and modeling technology on quadruped robot platforms. According to the company, this allows robots to map and interpret their surroundings in 3D for path planning, decision-making and task execution.
At the AWS showcase, XGRIDS presented what it described as a complete Real2Sim workflow, covering data capture, world model generation and simulation training.
The announcement presents this technology as infrastructure for physical AI applications, in which robots and other embodied AI systems are trained and tested in digital representations of real-world environments.
The company said its long-term focus is on building systems that convert real environments into machine-readable world models for AI training and reasoning. It linked this to use cases in sectors where robots operate in environments such as warehouses, cities and construction sites.
