Real2Sim Bridges Robots
At NVIDIA GTC, XGRIDS unveiled Real2Sim — a platform to train robots in digital twins that closely mirror real environments, aiming to cut sim‑to‑real gaps for logistics, healthcare and industrial automation. That’s a straight line to lower deployment risk and faster ROI in physical AI projects where sales cycles hinge on reliability and safety. (prnewswire.com)
XGRIDS announced that its spatial‑intelligence stack now supports NVIDIA Omniverse NuRec for OpenUSD‑based rendering, enabling export of reconstructed scenes into NVIDIA’s simulation formats. (prnewswire.com) At an AWS showcase during GTC, XGRIDS demonstrated a full “capture → world‑model → simulation” Real2Sim workflow that handoffs sensor captures into training environments for validation and policy training. (prnewswire.com) The Real2Sim pipeline shown in demos combines LiDAR and computer‑vision inputs for multimodal spatial perception and uses high‑fidelity 3D reconstruction to generate simulation‑ready world models. (prnewswire.com) NVIDIA’s Omniverse NuRec — introduced at SIGGRAPH 2025 — provides neural reconstruction and OpenUSD robot/sensor schemas and is integrated into Isaac Sim and CARLA to ingest camera and LiDAR data for sim‑to‑real workflows. (docs.nvidia.com) (roboticsbusinessnews.com) Company leadership listings identify Kaiyong Zhao as XGRIDS founder and CEO, and the firm describes production‑grade hardware‑software offerings for SLAM, real‑time reconstruction and spatial intelligence on its corporate site. (crunchbase.com) (xgrids.com) Conference coverage and independent reporting recorded live Real2Sim demonstrations that converted real indoor spaces into simulation environments for robot testing during NVIDIA’s GTC, highlighting the end‑to‑end capture and reconstruction steps. (rockingrobots.com) (prnewswire.com)
