Simulate every scenario
before your robot meets it

Rigyd generates SimReady objects and worlds each scenario requires, varied in shape, material, and physics, for your training and evaluation pipeline.

Simulation is only as good as its data

The physics is solved. The bottleneck is the data going in.

01

3D assets aren't simulation-ready

Most 3D assets lack the physical properties simulators need: mass, friction, collision geometry. Pretty geometry with no physics breaks grasping, stacking, and manipulation.

02

Environments don't exist at scale

Training robot policies takes thousands of diverse, physically accurate worlds. Building them by hand takes months and never covers the edge cases.

03

Variety doesn't scale by hand

Domain randomization demands endless variations of objects and scenes. Without scalable 3D data generation, the sim-to-real gap stays wide open.

Rigyd generates SimReady objects and physically accurate worlds on demand

We generate the 3D data your simulations need, objects with real physical properties, environments with collidable geometry, and endless variations for domain randomization and better sim-to-real transfer.

Step 01

Upload

Drop images / 3D models or describe what you need. We render each model from multiple camera views, and a vision model identifies surface materials and dimensions.

Step 02

AI estimates physics

Density, friction, and restitution come from engineering handbooks. Mass uses true mesh volume times material density; center of mass and inertia derive from geometry. Convex decomposition computes collision hulls from the mesh.

Step 03

Download SimReady USD and MJCF

Z-up, metric, single rigid bodies. USD Physics and MuJoCo MJCF exports, URDF coming soon. A 106-check validator confirms simulation-ready structure.

Output preview
ceramic_mug.usda 
# SimReady Output (OpenUSD)
def "CeramicMug" (
    prepend apiSchemas = ["PhysicsRigidBodyAPI", "PhysicsMassAPI"]
)
{
    float physics:mass = 0.34
    point3f physics:centerOfMass = (0, 0.052, 0)
    rel physics:simulationOwner = </World/PhysicsScene>

    def "CollisionMesh" (
        prepend apiSchemas = ["PhysicsCollisionAPI"]
    )
    {
        uniform token physics:approximation = "convexDecomposition"
        float physics:friction = 0.42
        float physics:restitution = 0.15
    }
}
ceramic_mug.xml 
<!-- SimReady Output (MJCF) -->
<mujoco model="ceramic_mug">
  <option gravity="0 0 -9.81" />

  <worldbody>
    <body name="CeramicMug" pos="0 0 0">
      <freejoint />
      <inertial pos="0 0.052 0" mass="0.34"
                diaginertia="0.0009 0.0009 0.0006" />

      <!-- convex-decomposition collider -->
      <geom name="collision" type="mesh" mesh="mug_hull"
            friction="0.42 0.005 0.0001" solref="0.02 0.15" />
    </body>
  </worldbody>

  <asset>
    <mesh name="mug_hull" file="mug_hull.obj" />
  </asset>
</mujoco>
Request API Access
// Build with Rigyd

Call it from anywhere you simulate

Terminal, REST, right inside Isaac Sim or MuJoCo. One API for all.

CLI & Python SDK

Convert from your terminal or Python. The result path goes to stdout — pipe it anywhere.

  • convert · generate · simulate · download
  • MuJoCo loader built in
CLI docs
zsh — robotics-project
# install + authenticate once
$ pip install rigyd
$ rigyd login
✓ key saved → ~/.config/rigyd/config.json

# 3D file → SimReady, optionally retopologized
$ rigyd convert chair.glb --tris 50000 --export isaac
  preprocessing → queued → running ▰▰▰▰▰▰ done
./assets/chair.usdz

# stdout is just the path — pipe it anywhere
$ blender $(rigyd convert scan.obj --export usd)

Compatible with

NVIDIA Omniverse platform logo, Rigyd outputs are compatible with NVIDIA Isaac Sim NVIDIA Isaac Sim
MuJoCo
Gazebo robotics simulator logo, Rigyd OpenUSD assets work with Gazebo and ROS 2
Unreal Engine logo, Rigyd simulation-ready assets are compatible with Unreal Engine

Built by a team that's already scaled 3D data.

We've built an end-to-end 3D platform trusted by Fortune 500 companies. In the last 12 months alone, we delivered 10 million immersive experiences. We know how to build the data infrastructure enterprises demand.

10M+
immersive experiences delivered
30K+
high-fidelity 3D assets generated

SOC 2 Type II

Enterprise-grade security with full audit trails and data protection from day one.

SSO & RBAC

Single sign-on, role-based permissions, and team management for any organization size.

Built to Scale

Generate millions of SimReady assets. From proof-of-concept to production data pipeline.

Our mission

Every machine will need to understand the physical world.

We're building the simulation data layer to make that future arrive faster.

Frequently asked questions

Everything you need to know about Rigyd and SimReady assets.

A SimReady asset is a 3D model enriched with the metadata and physical properties needed for accurate simulation. This includes rigid-body physics (mass, friction, restitution), collision meshes, semantic labels, and material specifications, all built on OpenUSD. Unlike a regular 3D model that only contains geometry and textures, a SimReady asset can be dropped directly into simulators like NVIDIA Isaac Sim, MuJoCo, or Gazebo and behave realistically. Rigyd automates this entire enrichment process using AI.

Rigyd accepts 3D models (.glb, .fbx, .obj), 2D images, and text descriptions. No pre-processing or format conversion is needed before upload. CAD formats are coming soon.

No. Rigyd handles everything automatically, file size optimization, mesh geometry cleanup, physics property estimation, collision mesh generation, and semantic labeling. Upload your raw 3D files and get simulation-ready OpenUSD output in minutes.

Domain randomization (DR) is the standard technique for crossing the sim-to-real gap: at training time you sample over a band of plausible physical and visual parameters, mass, friction, restitution, lighting, object placement, so the policy generalises across that variance instead of overfitting to a single simulator configuration. DR only works if the band you sample from spans what the policy will actually encounter in production. With a small fixed asset set you can randomise mass and friction per asset, but you cannot randomise object identity, geometry, or scene composition, the policy sees the same handful of objects every episode and learns their specific shapes. Rigyd makes the object and scene axes of DR practical: thousands of distinct physically accurate assets and full scenes generated at API rate, each with calibrated parameters (mass within 15-20% of measured, friction within ±0.1 coefficient) that DR can sample around. The GR00T N1 benchmark (NVIDIA, 2025) reports 40% better real-world performance specifically from physics-accurate, diverse synthetic training data, the diversity is what the lift comes from.

Rigyd's Enterprise plan includes API access for batch processing, so you can create entire scenes and objects programmatically. Request API Access to discuss your volume needs and get a custom plan.