Skip to main content

Closing the Sim-to-Real Gap: Training Humanoids in 3D Gaussian Environments | Niantic Spatial, Inc.

Hear more about how Niantic Spatial is working with NVIDIA and Flexion to close the Sim-to-Real gap! →

Closing the Sim-to-Real Gap: Training Humanoids in Real-World Environments

Scan the real world. Train robots that transfer.

Robots learn by failing, and failing in the real world is slow and expensive. Simulation fixes that, but only if the simulated world is a faithful stand-in for the real one. This session walks through an end-to-end real-to-sim-to-real workflow that makes it possible.

Featuring Eugene Chong (Niantic Spatial), Julian Nubert (Flexion), and Gavriel State (NVIDIA).

What you'll learn:

  • Create simulation-ready environments with 3D Gaussian splatting and collision meshes

  • Render Gaussian splat reconstructions with NVIDIA Omniverse NuRec in Isaac Sim

  • Train humanoid robot policies in NVIDIA Isaac Lab

  • Test robotics software in NVIDIA Isaac Sim

  • Build interoperable simulation workflows with OpenUSD

  • Close the sim-to-real gap for humanoid robotics