Industries · Robotics
Data for machines that move and manipulate.
Manipulation, grasping, and navigation policies have to generalize across articulation, contact, and clutter that is expensive to stage in the real world. Aurora generates that experience in simulation — including humanoid platforms.
Perception schematic
world / bin_picking_04
Sensor stack
RGBDepthSegmentation3D boxesObject pose
The challenge
- Rare grasps and contact events are hard to collect at scale
- Real-world clutter and articulation are costly to stage repeatedly
- Safety-critical failures cannot be provoked on real hardware
- Hand-labeling 3D pose and geometry is slow and error-prone
What Aurora provides
Coverage you cannot collect on the floor.
01
Randomized clutter, articulation, and object placement
02
Perfect 3D bounding boxes, object pose, and camera pose
03
Contact-rich manipulation scenarios swept systematically
04
Humanoid and arm platforms with configurable sensor rigs
Illustrative scenario
bin_picking_04
Cluttered bin picking swept across 50k object arrangements and lighting conditions.
Other industries
Get started
Bring Aurora to your robotics stack.
Tell us about the machines you are training and the environments they operate in. We will scope a synthetic-data plan with you.