Train Your Robot Policy in the World It Will Run In
Capture with a phone, rig, or drone. Turn real-world scans into high-fidelity digital twins: simulation-ready Gaussian splats with aligned meshes. Train and run evaluations inside them, then deploy with reliable localization in any environment.
Tackle the fundamental challenges that limit real-world deployment
Policies trained only in synthetic environments struggle in messy, irregular real-world settings. With USDZ export in Scaniverse, a single 360 capture becomes a simulation-ready digital twin – a Gaussian splat paired with an aligned mesh – so a robot can train in a faithful replica of its destination and arrive ready to work from day one.
From Real Capture to Simulation-Ready Training
The USDZ export in Scaniverse turns a five-minute 360 capture into a simulation-ready environment for NVIDIA Isaac Sim and Isaac Lab; a Gaussian splat and an aligned mesh in a single file.
The splat carries the real-world lighting, texture, and clutter that synthetic scenes approximate or omit, so a robot’s vision system can train and be evaluated against how its destination actually looks. The mesh, derived from the same capture, supplies the collision geometry it navigates.
Together, they let teams fine-tune policies for a specific environment and test how those policies are likely to perform before deployment.
Niantic Spatial Partners with Coco Robotics to Accelerate the Future of Autonomous Delivery
Niantic Spatial will be a core infrastructure partner for Coco, deploying spatial AI and its Visual Positioning System (VPS) to further enhance the company’s advanced robot delivery fleet.
Building the Spatial Infrastructure for General-Purpose Robotics
By Kit Gilbert, Head of Strategic Partnerships
Robots have long been part of industrial settings, but they've traditionally been narrow, task-specific systems. The scope is shifting, driven by advances in hardware and AI foundation models. The goal is to build robots that can operate flexibly in complex, human-scale environments, not just repeat fixed tasks. A key challenge standing in the way is navigation in those human environments.