Percept — Foundation Model for the Physical World

Technology

From raw capture to real understanding.

Percept's foundation model processes any sensor input through three stages. Each stage builds on the last. The output is a structured, queryable, simulatable representation of the physical world.

Reconstruct

Turn any video, LiDAR scan, or satellite image into a metrically accurate 3D model. Sub-centimeter precision, real-world coordinates, in minutes not hours.

Understand

Every object, material, defect, spatial relationship — encoded as a structured, queryable graph. Not pixels. A representation engineers and AI agents can reason over.

Click a node to inspect its properties, connections, and metadata.

LEGEND

CORE

ENTITY

PROPERTY

RELATION

ALL NODES

Scene

Building

Terrain

Vehicle

Vegetation

Road Network

Sensor Origin

Physics Layer

Semantic Labels

Spatial Index

Temporal State

Adjacency

Occlusion

Simulate

Fire spread, structural failure, flood inundation — run on actual geometry and real terrain. Physics that reflects the world as it is, not a generalized approximation.

Raw sensor data becomes a metric 3D scene graph — objects, terrain, defects, and risk zones structured for query and simulation.

Products

Two products. One pipeline.

The same Reconstruct → Understand → Simulate foundation, delivered in two forms: a visual operating environment for operators, and a programmatic API for builders.

for operators

SpatialOS

Public Safety · Utilities · Construction · Transportation

› Find critical defects within 5m of beam_047

12 results3 critical9 ok

Emergency managers, inspectors, city planners — upload footage, query the scene graph, run simulations. No code required.

Upload any capture"Video, LiDAR, satellite"

Query the scene graph"Find defects within 5m of beam_047"

Simulate scenarios"Fire spread, NW 12.5 km/h, 4 hours"

Explore SpatialOS →

for builders

Percept API

Drone · Robotics · Autonomous Systems · Edge AI

percept_sdk.py

import percept

# Reconstruct from any sensor
scene = percept.reconstruct(
  source="drone_capture.mp4",
  mode="metric_3d"
)

# Query the scene graph
defects = scene.query(
  type="structural_defect",
  severity="critical"
)

Drone manufacturers, robotics platforms, and autonomous systems. Three endpoints. One SDK.

percept.reconstruct()scene.query()scene.simulate()

Read the Docs →

Deployments

Proven in the field.

From earthquake response to bridge inspection. Real deployments. Real outcomes.

Emergency

Turkey Earthquake Response

47 structuresassessed in 3 hours

Infrastructure

Pacific NW Bridge Inspection

12 critical defectsidentified across 4 spans

Wildfire

CA Fire Spread Prediction

4-houradvance warning accuracy

The intelligence layer for
everything physical.

We started
in the field.

From raw capture to real understanding.

Reconstruct

Understand

Simulate

Two products. One pipeline.

SpatialOS

Percept API

Proven in the field.

Turkey Earthquake Response

Pacific NW Bridge Inspection

CA Fire Spread Prediction

Built for the industries that build and protect the world.

Articles, notes, and field reports.

The intelligence layer foreverything physical.

We startedin the field.

From raw capture to real understanding.

Reconstruct

Understand

Simulate

Two products. One pipeline.

SpatialOS

Percept API

Proven in the field.

Turkey Earthquake Response

Pacific NW Bridge Inspection

CA Fire Spread Prediction

Built for the industries that build and protect the world.

Articles, notes, and field reports.

The intelligence layer for
everything physical.

We started
in the field.