/PRNewswire/ — For robots to operate reliably in the real world, they must train in environments that accurately represent it. This is the central challenge of the Real2Sim method, and the main reason for the presence of XGRIDS at the GTC2026 conference (March 16-19 in San Jose).
image
XGRIDS’ spatial intelligence solution now supports NVIDIA Omniverse NuRec to render based on OpenUSD, and has been featured multiple times: startup pitch deck, robotics demos in the NVIDIA ecosystem, and joint presentation with Amazon Web Services (AWS).
Real2Sim: from physical space to AI training
During the startup pitch presentation, XGRIDS Director SunnyLiao asked a central question:
How can we create training environments that truly reflect real-world conditions?
XGRIDS responds to this with its R&D portfolio built on concrete data. By combining laser remote sensing (LiDAR) and computer vision for multimodal spatial perception with high-fidelity 3D reconstruction, it is possible to transform physical environments into models of the world usable for simulation.
Compared to manual 3D modeling, this approach:
- Reduces the cost of creating high-fidelity environments
- Allows for continuous updates as real spaces evolve
- Ensures closer proximity between simulation and real-world deployment
Developers at the GTC conference noted that this method provides a more practical route to training robots and validating robotic solutions.
Spatial intelligence at the service of physical AI
Beyond this initial stage, XGRIDS solutions could be observed in numerous exhibition areas of the GTC conference.
At NVIDIA’s robotics solutions event, XGRIDS demonstrated how its solution for embodied AI systems works. By deploying spatial perception and modeling on quadruped robotic platforms, robots can continuously map and understand their environment based on a full 3D spatial structure for path planning, behavioral decision-making and task execution, instead of relying solely on local sensors for immediate obstacle avoidance.
This solution demonstrated how spatial intelligence integrated into embodied AI systems to enable robots to understand full 3D environments, and thus plan their path, make decisions and execute tasks.
During the AWS demo, XGRIDS demonstrated a complete Real2Sim workflow, from capture to simulation training to world model generation.
Perspectives
The long-term goal of XGRIDS remains constant: to build a spatial intelligence infrastructure that converts real-world environments into models of the world that AI systems can understand, reason about and train on.
The GTC2026 conference marked a milestone in the integration of this work into the physical AI ecosystem. As embodied AI systems move out of labs and into warehouses, cities, and construction sites, the demand for accurate, scalable digital representations of environments will continue to grow.
XGRIDS builds the capture-simulation layer that makes this progression possible.
Media Relations:
[emailprotected]
https://www.xgrids.com
Photo – https://mma.prnewswire.com/media/2938856/image.jpg
