Skip to content
Aurora

Aurora Forge

In development

A programmable world generator for Physical AI.

Aurora Forge is the first Aurora product — a programmable synthetic-data layer being built for robotics and autonomous systems. You define a world once; Forge generates the labeled, multi-modal data your models need to learn it.

Request AccessSee the fleet

The workflow

01

Define

Describe the world, its sensors, and the rules for how it should vary.

02

Simulate

Physics, materials, and sensors are evaluated across thousands of controlled scenarios.

03

Generate

Observations stream off GPU workers with ground truth attached to every frame.

04

Train

A sharded, versioned dataset lands ready for your training and evaluation pipelines.

What you define

A world definition is a program, not a fixed scene. Every element can be a distribution instead of a constant.

  • Environments
  • Objects
  • Robots
  • Articulated systems
  • Cameras
  • Sensors
  • Materials
  • Lighting
  • Weather
  • Object & robot placement
  • People & trajectories
  • Occlusion
  • Edge cases
  • Randomization rules
What comes out

Each scenario yields synchronized sensor observations plus perfect, automatically produced ground truth.

  • RGB
  • Synthetic video
  • Depth
  • LiDAR
  • Radar
  • Optical flow
  • Surface normals
  • Semantic segmentation
  • Instance segmentation
  • 2D boxes
  • 3D boxes
  • Object pose
  • Camera pose
  • Sensor metadata

Concept API

Generation as a single call.

The interface below is an illustrative sketch of the Forge SDK — it shows the intended shape of the API, not a released public product. A job describes a world, its sensors, a scenario budget, and what to randomize.

Explore the developer model further on the Technology page.

forge.ts — illustrative
1const dataset = await aurora.generate({2  world: 'warehouse_07',3  sensors: ['rgb', 'depth', 'lidar'],4  scenarios: 1_000_000,5  randomize: ['lighting', 'materials', 'objects']6})78dataset.status     // 'generating'9dataset.frames     // stream of labeled frames10dataset.annotations// ground truth per frame

Aurora Forge

Define a world. Generate its data.

We are building Forge with a small group of robotics and perception teams. Tell us what you are working on and we will be in touch.

Request AccessTalk to Aurora