Humanoids Summit Silicon Valley 2026
- Dates
- December 1-2, 2026
- Location
- San Mateo Event Center
Visit Niantic Spatial at Booth #50
Speakers
Agenda
1 sessionFrom Simulation to Real World: Building the Infrastructure Embodied AI Needs to Work Anywhere
Embodied AI policies trained only in synthetic environments struggle when they meet the details of a real deployment site: uneven floors, clutter, changing layouts, and transparent surfaces. Closing that sim-to-real gap costs time, money, and customer trust — one reason robotics POCs often run six to twelve months before a purchase order, with deals stalling on doubt that a policy will hold in a particular environment.
The answer is to combine the scale of synthetic data with the accuracy of real-world environments: reconstructions detailed enough that what a policy sees is what it collides with, delivered in formats that drop straight into simulators like NVIDIA Isaac Sim and Isaac Lab.
Kit Gilbert, GM of Embodied AI at Niantic Spatial, presents the company's real-to-sim pipeline, built and evaluated with partners Flexion Robotics and NVIDIA. Trained on a reconstruction, an RGB navigation policy ran zero-shot on the real robot and handled cases where a depth-input policy failed - arriving with evidence it already works in that exact environment, rather than doubt.


