KYROS
Training Curricula

The sequence matters as much as the environment.

Training a robot on the wrong sequence of tasks produces a robot that only works in the lab. Kyros curricula are ordered difficulty progressions, designed by researchers who have closed the sim-to-real loop.

Anatomy of a curriculum.

STAGE 01

Single Object, Controlled Placement

Robot encounters one object in a fixed position. Task: grasp and move. Zero variation. Build the motor primitive.

success threshold: 90%
STAGE 02

Multi-Object, Fixed Positions

Three objects. Fixed. Robot learns to select the correct target and ignore distractors.

success threshold: 85%
STAGE 03

Single Object, Random Placement

Object appears anywhere in the workspace. Domain randomization begins. Robot learns to generalize position.

success threshold: 82%
STAGE 04

Multi-Object, Random Placement + Clutter

Three objects, randomized positions, with distractor clutter. Robot must isolate, plan, and execute under noise.

success threshold: 78%
STAGE 05

Full Domain Randomization

Lighting, materials, friction, mass, camera pose — all randomized. The policy that survives this is the policy that transfers.

success threshold: 70%
20+
Pre-built curricula
Forkable
Adapt any curriculum to your robot
Researcher-designed
Built by sim-to-real practitioners

Train in the right order.

Get early access to the full curriculum library.