The operating layer for robot training.

Kyros packages simulation worlds, task definitions, training runs, and transfer evaluation into one production pipeline. Less scene-building. More policies that survive contact with reality.

kyros_runtimeenvironment → rollout → transfer
sdk pipelineeval run
seedscoreexport
physics
policy
transfer

Environment system

Versioned worlds with geometry, contact dynamics, sensors, objects, and randomization configs managed together.

Training runtime

Vectorized rollouts, curricula, reproducible seeds, hosted runs, and policy export for modern robotics teams.

Transfer evaluation

Benchmarks, reports, edge-case suites, and deployment confidence signals before your hardware takes risk.

500+
environment templates across household, industrial, medical, and outdoor robotics.
2,000+
task definitions with rewards, success criteria, observations, and curricula.
10k+
procedural variations per high-priority environment class.

Browse the environment library.

Kitchen — Residential
Manipulation Verified
Load Environment
Warehouse — Fulfillment
Logistics Verified
Load Environment
Hospital Room
Healthcare Verified
Load Environment
Outdoor Terrain
Locomotion Verified
Load Environment
Assembly Line
Manufacturing
Load Environment
Home Office
Manipulation
Load Environment
Construction Site
Outdoor
Load Environment
Laboratory
Precision
Load Environment
Retail Store
Service
Load Environment

2,000+ training tasks. Ready to use.

pick_and_placedoor_openingobject_sortingbin_pickingstair_climbingbalance_recoverylegged_locomotionnavigationcloth_foldingliquid_pouringassembly_pegscrew_insertionobstacle_avoidancehuman_followinghandover_tasktool_use

Integrate in minutes.

API Reference
GET/v1/environments
List all environments
POST/v1/environments/load
Load an environment by id
POST/v1/training/start
Begin a training run
GET/v1/policies/:id
Retrieve a trained policy
Python 3.11
# Quick start
import kyros
# Authenticate
kyros.auth(api_key="kyr_...")
# List environments
envs = kyros.environments.list(category="manipulation")
# Train and export
env = kyros.environments.load("kitchen_v4.2")
policy = kyros.train(env, timesteps=1e7)
policy.export("model.pt")

Start building today.

Join hundreds of robotics teams already training in simulation.