Pipeline demo
Community Asset Program: vote on what we build nextView the challenge
Simulation infrastructure for deployment teams
Bring the real world into simulation.
Dirac turns camera video of real sites and objects into physics-accurate simulation, so robots can train in the environments where they will actually work.
Camera video
Physics-ready scene
- 01Capture
- 02Reconstruct
- 03Resolve physics
- 04Simulate
Real2Sim
Your site, rebuilt for simulation.
Walk the space once with a camera. Dirac reconstructs its geometry and resolves the physical properties that make the scene useful for training: scale, mass, inertia, friction, and joints.
The result is a simulation-ready environment built from the place where the robot will actually run, instead of a generic stand-in assembled by hand.
How the pipeline works- Input
- Single-camera video
- Output
- USD, Isaac Sim ready
- Physics
- Predicted, confidence stated
- Turnaround
- Hours, not weeks
00:51
Reconstructed scene
Inspect the delivered environment
Explore the scene
Loads the full 3D reconstruction on demand.
Simulation assets
Objects that behave like the real thing.
A mesh can look right and still teach the wrong behavior. Dirac assets combine photoreal geometry with mass, inertia, friction, joint behavior, confidence, and real-world validation.
- Geometry
- Deployment-specific
- Physics
- Automatically resolved
- Confidence
- Stated per value
- Evidence
- Real-world validation
Purple chairDrag to rotate, or use the arrows
Coming soon
The rest of the deployment loop.
A scene is the beginning. We are building the infrastructure that tests policies against the real site and keeps them current as that site changes.
Why Dirac
Specificity is what makes robots deployable.
Site-specific
Built from the actual room, objects, and operating conditions your robot will encounter.
Physics-aware
Mass, inertia, friction, and joint dynamics matter as much as geometry and appearance.
Evidence-backed
Values carry stated confidence and are checked against real observations instead of hidden assumptions.
The team
Founders from CMU, Microsoft, and frontier AI research labs, with experience building autonomous driving perception, GitHub Copilot, and 3D computer vision systems. We are building the physics layer that every robot will learn from.
Get started
Show us where your robot needs to work.
Send us the site and the task. We will tell you what we can reconstruct and how fast.