Niantic Spatial, Flexion, and NVIDIA: Closing the Sim2Real Gap for Humanoids
Niantic Spatial, Inc.
Niantic Spatial, Flexion, and NVIDIA successfully demonstrate a pipeline that trains RGB-based humanoid navigation policies in simulation and transfers them zero-shot to reality.Summary
Niantic Spatial, Flexion, and NVIDIA have introduced an end-to-end pipeline that bridges the sim2real gap for humanoid robots by utilizing high-fidelity digital twins. By combining Niantic Spatial's real2sim capture technology—which generates photorealistic 3D Gaussian splats and aligned collision meshes from standard 360° video—with NVIDIA's Isaac Sim and Flexion's reinforcement learning frameworks, developers can train robust RGB-based navigation policies. This approach enables robots to learn complex traversal and obstacle avoidance in simulated environments before achieving successful zero-shot transfer in real-world offices.
(Source：Niantic Spatial, Inc.)