其域创新 XGRIDS 在 NVIDIA GTC 2026 展示 Real2Sim 空间智能方案
XGRIDS at NVIDIA GTC 2026: Bridging Real-World Spaces and Physical AI Through Real2Sim
其域创新(XGRIDS)在 NVIDIA GTC 2026 上展示 Real2Sim 方案,通过 LiDAR 与计算机视觉的多模态空间感知及高保真三维重建,将真实环境转为可用于仿真训练的世界模型。
- Home
- Insights
- Events & Webinars
- XGRIDS at NVIDIA GTC 2026: Bridging Real-World Spaces and Physical AI Through Real2Sim
XGRIDS at NVIDIA GTC 2026: Bridging Real-World Spaces and Physical AI Through Real2Sim
For robots to operate reliably in the real world, they need to train in environments that accurately represent it. That's the core problem behind Real2Sim — converting physical spaces into simulation-ready 3D environments — and it's what XGRIDS presented at NVIDIA GTC 2026 (March 16–19, San Jose).
XGRIDS' spatial intelligence solution now supports NVIDIA Omniverse NuRec for OpenUSD-based rendering and appeared across multiple GTC venues: a featured presentation during Startup Pitch, a solution showcase as part of NVIDIA’s Robotics demonstrations, and a joint demonstration with Amazon Web Services (AWS).
Startup Pitch: Real2Sim for Physical AI
XGRIDS Director of Global Sales & Marketing Sunny Liao presented the company's Real2Sim solution during Startup Pitch, centered on a specific question: when intelligent agents need to operate in physical environments, how do you build training environments that faithfully represent reality?

XGRIDS' answer is a pipeline driven by real-world data. Through multimodal spatial perception — combining LiDAR with computer vision — and high-fidelity 3D reconstruction, real environments are converted into world models usable for simulation training.

Compared to traditional manual 3D modeling, this pipeline significantly reduces the cost of building high-fidelity training scenes and allows environments to be continuously updated as physical spaces change — keeping training conditions closer to actual deployment.
In post-session conversations, developers and ecosystem partners noted that generating simulation environments directly from real-world spatial captures offers a more practical path for robot training and system validation than manual scene construction.
Across the Show Floor: Spatial Intelligence for Embodied AI
Beyond the Inception stage, XGRIDS' solutions appeared at multiple showcase areas across GTC.

At the NVIDIA Robotics session, XGRIDS demonstrated its solution for embodied AI systems. By deploying spatial perception and modeling on quadruped robot platforms, robots can continuously map and understand their environments, using full 3D spatial structure for path planning, behavioral decision-making, and task execution rather than relying solely on local sensors for immediate obstacle avoidance.

At the AWS partner showcase, XGRIDS presented its Real2Sim pipeline at a systems level — showing how real-world spatial data enters AI training workflows to form a complete loop from environment capture, to world model construction, to simulation training.

Looking Ahead: Building the Capture-to-Simulation Layer
XGRIDS' long-term focus remains consistent: building spatial intelligence infrastructure that converts real-world environments into world models AI systems can understand, reason about, and train on.
GTC 2026 marked a step in bringing this work into the international Physical AI ecosystem, including through collaboration with AWS. As embodied AI systems move from labs into warehouses, cities, and construction sites, the demand for accurate, scalable digital representations of those environments will only grow.
XGRIDS is building the capture-to-simulation layer that makes that possible.
来源:其域创新 XGRIDS · xgrids.com