Real2Sim · Sim2Real

Scan the floor once. Train the robot inside it.

One walkthrough of your facility becomes a simulation-ready digital twin in NVIDIA Isaac Sim: photoreal appearance plus collision geometry, at metric scale. Robots train there in parallel, then deploy to the real floor on the same map they learned in.

Bring us a floor Maryland M4.0: up to 75% funded

NVIDIA Isaac Sim · autonomous tour of a warehouse twin · 5 of 5 waypoints reached

The pipeline

Four steps from a walkthrough to a robot that knows the building.

The same payload that captures a building for BuiltLedger captures it for simulation. Nothing is modeled by hand.

01 · CAPTURE

One pass with the payload

LiDAR, 360° video, and depth from the payload on a quadruped. No operator on the floor, no production stop. A floor plan works as a starting point when a scan hasn't happened yet.

02 · RECONSTRUCT

Splat for looks, mesh for physics

The capture becomes a Gaussian splat for photoreal appearance and a collision mesh for contact. Both are packaged as OpenUSD, gravity-aligned and at metric scale.

03 · TRAIN

Thousands of attempts, in parallel

The twin loads into Isaac Sim and Isaac Lab. Dozens of copies of the robot learn navigation and inspection routes at once, failing safely where it costs nothing.

04 · DEPLOY

Same map on the real floor

The policy ships to the NVIDIA Jetson on the robot. The scan it trained in is the map it localizes against, so day one on site looks like day one thousand in sim.

What the twin is made of

Our own office, rebuilt as a training ground.

909 Rose Avenue, 4th floor. The plan is segmented into envelope, walls, core, furniture, and free space, then exported as an Isaac USD scene with box colliders and a navigable corridor network. Each of the 67 suites gets a goal point outside its door.

Segmented floor plan of the 909 Rose Avenue 4th floor: walls, core, furniture, the green navigable corridor network, and 67 numbered suite goal points — Segmented layers and the navigable corridor network, with a goal point outside every suite.

The same floor as an Isaac Sim collision scene, ready for parallel training runs.

From the runs

Numbers from runs we can replay for you.

5 / 5

Waypoints reached on the autonomous warehouse tour in Isaac Sim

603,920

Vertices in the twin the robot captured on that tour

Robots training at once in the 909 Rose office twin, on one GPU

20,307 sf

Office floorplate rebuilt from a plan and calibrated to the inch

On the real floor

The other half of the loop runs on legs.

The payload rides a Unitree Go2 with a Jetson Orin NX. Autonomous navigation on real floors and stairs, and a semantic map of the building built on board as it walks.

Autonomous capture inside a live commercial building. No operator on the floor.

Stairs and tight interiors, unassisted. Built for the buildings people actually work in.

On-board vision builds a semantic 3D map as it walks: rooms segmented, equipment and assets labeled in real time.

For manufacturers

Get the plant ready for robots before the robots arrive.

A twin of your floor is the cheapest place to find out where an AMR gets stuck, which inspection route a quadruped can actually walk, and how a workflow changes when a cell moves. Train it in the twin, then bring it in. Maryland manufacturers can fund the capture and the hardware through Manufacturing 4.0.

See the M4.0 sheet

Robotic inspection routes proven in sim before a robot touches the floor
Workflow and cell layout changes tested against the real geometry
Asset registry and nameplate data captured on the same pass
Twin stays current with every scan, so the training ground never goes stale

Where the pipeline stands

What ships today, and what we're closing next.

We'd rather show you a run than a render. Here is the honest state of the loop.

Running today

Autonomous capture on real floors and stairsUnitree Go2 · Jetson Orin NX · LiDAR + 360° + depth
Twin export to OpenUSD for NVIDIA Isaac SimWarehouse twin from a robot tour · 603,920 vertices
Autonomous navigation in the twin5 of 5 waypoints on the warehouse tour
Floor plan to Isaac Lab training scene909 Rose L4 · 67 goals · 64 parallel environments

Closing next

Learned navigation policy deployed zero-shot to the physical Go2Office-nav training runs are live; the corridor walker is being retrained
Photoreal splat capture from a synced 360° camera ringGlobal-shutter rig on the Orin NX, in build
Inspection behaviors trained in the twinNameplate and thermal checks as trained routes