Reconstruct
Generate accurate, geo-referenced 3D digital twins from a variety of input data types. Designed for high fidelity and efficiency, running seamlessly from edge devices to the cloud for real-time mapping, localization, and visualization.
Reconstruct integrates diverse data sources from Niantic Spatial Capture↗ and third party capture solutions into seamless, high-fidelity digital twins ready for mapping and analysis.
Processing Pipeline
Scan Processing
Clean, filter, and normalize raw sensor inputs from drones, robots, and handheld systems.
Registration
Align and merge multi-pass scans to ensure centimeter-level spatial accuracy.
Geo-Referencing
Anchor models to real-world coordinate systems for interoperability with GIS, simulation, and autonomy platforms.
Reconstruction Modes
On-Device 3D Reconstruction
-
Process 3D data directly on iOS or Android. Fast, private, and mobile.
-
Generate textured meshes or Gaussian splats in real time, fully offline.
-
Ideal for edge computing, field capture, and AR experiences with no cloud dependency.
Cloud-Based 3D Reconstruction
-
Deliver high-fidelity, geo-referenced 3D models in the cloud.
-
Generate photorealistic meshes and splats enhanced by semantic understanding.
-
Ideal for large-scale mapping, multi-user projects, and enterprise digital twins.
Best experienced on desktop
Splat Viewer
This demo showcases how Niantic Spatial transforms high-fidelity 3D models into interactive canvases for real-world workflows. Exploring the Gaussian splat reveals content spatially anchored to the location.
Built for the Real World
Defense
Enable mission-critical capabilities for spatial autonomy through intelligent location surveying, high-fidelity 3D reconstruction, and immersive AR/VR environments.
Robotics & Autonomy
Generate high resolution 3D environments optimized for perception, path planning, and navigation across dynamic, real-world conditions.
Intelligent Field Operations
Deliver detailed geospatial context for infrastructure inspection, maintenance, and remote operations with scalable, continuously updated digital twins.
Frequently Asked Questions
What is 3D reconstruction?
+
3D reconstruction turns images, video, or other sensor data into a three-dimensional representation of a real object or place. Niantic Spatial Reconstruct generates high-fidelity, geo-referenced digital twins that can support mapping, localization, visualization, analysis, simulation, and real-world operations. The result is more than a visual model: it is a spatially anchored representation designed to connect physical environments with the systems that use them.
How does 3D reconstruction work?
+
A 3D reconstruction pipeline processes raw sensor inputs, aligns multiple scans, and anchors the resulting model to a real-world coordinate system. Reconstruct describes these stages as scan processing, registration, and geo-referencing. Niantic Spatial's broader technical explanation adds that reconstruction also requires estimating camera positions and recovering scene geometry, so the output is physically grounded rather than only visually plausible.
What data can be used for 3D reconstruction?
+
3D reconstruction can use imagery and sensor data from multiple capture sources, depending on the project. Reconstruct is designed to integrate Niantic Spatial Capture data and third-party solutions, including inputs from drones, robots, and handheld systems. Niantic Spatial's broader platform context also describes workflows using smartphones, 360° cameras, satellite imagery, LiDAR, and other sensors, enabling teams to work with the data they already have.
What is the difference between on-device and cloud-based 3D reconstruction?
+
On-device reconstruction processes 3D data directly on iOS or Android devices, enabling fast, private, offline capture and real-time generation of textured meshes or Gaussian splats. Cloud-based reconstruction delivers high-fidelity, geo-referenced models for large-scale mapping, multi-user projects, and enterprise digital twins. The right mode depends on whether the priority is field mobility and offline operation or scale, collaboration, and centralized processing.
What are 3D reconstructions used for?
+
3D reconstructions are used to create shared spatial context for mapping, visualization, localization, simulation, inspection, maintenance, and remote operations. Reconstruct highlights defense, robotics and autonomy, and intelligent field operations. Across Niantic Spatial's broader work, geometrically accurate reconstructions also support infrastructure assessment, energy and industrial site analysis, and robotics training in environments that reflect the real place where systems will operate.
What is the difference between a 3D mesh and a Gaussian splat?
+
A 3D mesh represents a scene through connected geometric surfaces, while a Gaussian splat represents it through many viewable 3D Gaussian elements that can produce photorealistic imagery. Both can be outputs of a reconstruction workflow, and the best choice depends on the downstream task. Meshes are useful for simulation and training, while aligned Gaussian splats provide immersive visual representations for exploration and visualization.