ROBOTICS

We Build the Worlds
Your Robots Learn From

Auto-generate physics-complete High-fidelity, High-quality SimReady data

Decades of CAD into sim-ready files —

every part tree preserved, physics & joints injected.

OVERVIEW
OVERVIEW

Why TRINIX?

Why TRINIX?

Since it emerged, Physical AI has drawn more attention than ever and teaching robots to perform tasks reliably has become critical. But robot learning demands high-quality, diverse, and massive training data, and that data is in short supply. Each training asset must be realistic 3D data that also carries every physical property materials, mass, friction, joints so producing even one takes a long time. This is where TRINIX comes in.

Since it emerged, Physical AI has drawn more attention than ever and teaching robots to perform tasks reliably has become critical. But robot learning demands high-quality, diverse, and massive training data, and that data is in short supply. Each training asset must be realistic 3D data that also carries every physical property materials, mass, friction, joints so producing even one takes a long time. This is where TRINIX comes in.

PROCESS
PROCESS

How It Works

Step 01

Generate 3D CAD
from text or an image

TRINIX generated 3D model in CAD simply with prompt.

Step 02

Inject physics, joints,

and materials automatically

TRINIX generated 3D model in CAD simply with prompt.

Step 03

Separate visual and collision

mesh to ensure smooth simulation

TRINIX generated 3D model in CAD simply with prompt.

Step 04

Scale into 1,000+

randomized variations

TRINIX generated 3D model in CAD simply with prompt.

Step 05

Export as Simulation ready data for Isaac Sim or Mujoco

TRINIX generated 3D model in CAD simply with prompt.

Convert Your CAD without Loss

Convert Your CAD without Loss

01. Upload Your CAD

Bring in the CAD data you already own.

02. Describe What Matters

Tell us which details need to be preserved.

03. Part-Level Reconstruction

TRINIX analyzes the data and rebuilds the part composition.

04. Automatic Physics Injection

Physics and articulation data are inferred and injected automatically.

05. Collision Mesh Separation

Visual and collision mesh are separated for smooth simulation.

06. Simulation-Ready Export

Export into a simulation-ready file.

01. Upload Your CAD

Bring in the CAD data you already own.

02. Describe What Matters

Tell us which details need to be preserved.

03. Part-Level Reconstruction

TRINIX analyzes the data and rebuilds the part composition.

04. Automatic Physics Injection

Physics and articulation data are inferred and injected automatically.

05. Collision Mesh Separation

Visual and collision mesh are separated for smooth simulation.

06. Simulation-Ready Export

Export into a simulation-ready file.

SOLUTION
SOLUTION

From Asset to Simulation

Available
with Articulations

Available
with Articulations

Auto-defined

articulation

Auto-defined

articulation

Doors, drawers, and hinges get their joints and drive axes defined automatically.

Doors, drawers, and hinges get their joints and drive axes defined automatically.

Fine-tune

with parameters

Fine-tune

with parameters

Adjust motion range, axes, and limits with numeric parameters.

Adjust motion range, axes, and limits with numeric parameters.

Available
with Deformables

Available
with Deformables

Inferred

deformable motion

Auto-defined

articulation

Inferred

deformable motion

Even the hard-to-predict motion of cables, cloth, and sacks is reproduced through inference.

Even the hard-to-predict motion of cables, cloth, and sacks is reproduced through inference.

Fine-tune

with parameters

Fine-tune

with parameters

Adjust stiffness, friction, and more with numeric parameters.

Adjust stiffness, friction, and more with numeric parameters.

black flat screen computer monitor

1,000+
Variations

1,000+
Variations

From one generated model,

1,000+ variations

From one generated model,

1,000+ variations

In simulation, 10,000+ more variations for training

In simulation, 10,000+ more variations for training

Domain Randomization
(VLA Data)

Domain Randomization
(VLA Data)

Placeholder

Placeholder

Placeholder

Placeholder

black flat screen computer monitor
USE CASES
USE CASES

How TRINIX is Used

  • Grasping-training data

    for robot arms

  • Warehouse

    logistics-robot simulation

  • Manipulation-training

    environments for humanoids

  • Grasping-training data

    for robot arms

  • Warehouse

    logistics-robot simulation

  • Manipulation-training

    environments for humanoids

OTHER SOLUTIONS
OTHER SOLUTIONS

Explore Other Solutions

man in black shirt standing in front of woman in black shirt

Digital Twin

Build and validate factories and new products in simulation first.

close-up photography of black metal gears

Industrial Design

Free yourself from stroke-by-stroke modeling. Automate the process instead.

black flat screen computer monitor

Asset Library

Browse a library of SimReady assets, ready to drop into your simulation.

CONTACT US
CONTACT US

Start Building the Data
Your Robots Learn From

By clicking submit, you agree to our Privacy Policy.

By clicking submit, you agree to our Privacy Policy.

[Our Locations]

Korea HQ

18F, Kins Tower, 8 Seongnam-daero 331beon-gil, Bundang-gu, Seongnam-si, Gyeonggi-do, Republic of Korea

R&D Center

Office 403, Global Biz Center, 43 Changeop-ro, Sujeong-gu, Seongnam-si, Gyeonggi-do, Republic of Korea

US Office

3003 N 1st St., San Jose, CA 95134, United States

ⓒ 2026 NDotLight Co., Ltd.

Korea HQ

18F, Kins Tower, 8

Seongnam-daero 331,

Bundang-gu,

Seongnam-si,

Gyeonggi-do,

Republic of Korea

R&D Center

Office 403,

Global Biz Center, 43 Changeop-ro,

Sujeong-gu,

Seongnam-si,

Gyeonggi-do,

Republic of Korea

US Office

3003 N 1st St.,

San Jose,

CA 95134,

United States

ⓒ 2026 NDotLight Co., Ltd.