KYROS
Simulation infrastructure for physical AI

Train robots inthe world they'llactually live in.

Kyros provides physics-accurate simulation environments where robotics teams train AI models at scale, before a single real-world deployment.

430,000x
faster than real time
10,000+
environment variations
Zero-shot
sim-to-real transfer

Robots learn the hard way.

Until now.

Real-world training

  • Weeks of setup per task
  • Robot damage and safety risk
  • Impossible to scale variations
  • No edge case coverage

Kyros simulation

  • Environments ready in minutes
  • Zero hardware risk
  • 10,000+ procedural variations
  • Full edge case library
0%
Average performance drop from lab to real world
0 month
Engineering time to build a single simulation environment from scratch
$0K
Loaded engineer cost saved per environment reused
Platform

From prompt to policy.

Three steps. Zero hardware required.

01

Choose an Environment

Browse our marketplace of pre-built, physics-accurate simulation environments. Kitchen, warehouse, outdoor terrain, hospital room. Every major task category is covered.

genesis_powered
02

Configure Your Training

Select a task from our library, set your domain randomization parameters, choose your curriculum difficulty progression. No environment-building required.

10,000+ variations
03

Deploy to Real Hardware

Export your trained policy. Our transfer-optimized environments are validated for real-world deployment. What trains in Kyros, transfers to your robot.

zero-shot transfer
Python 3.11
# Kyros Platform SDK — v1.0
import kyros
# Load a pre-built kitchen environment
env = kyros.environments.load("kitchen_v4.2",
robot="franka_panda",
variations=10000,
domain_randomization=True
)
# Configure training curriculum
curriculum = kyros.curriculum.build(
task="pick_and_place",
difficulty="progressive",
edge_cases=["clutter", "occlusion", "low_light"]
)
# Train policy — runs 430,000x faster than real time
policy = kyros.train(env, curriculum, timesteps=1e8)
# Export for real-world deployment
policy.export("my_robot_policy.pt")
# Ready to deploy. Zero real-world training required.
Features

Everything your sim-to-real pipeline needs.

Environment Marketplace

Hundreds of pre-built, task-specific environments. Kitchens, warehouses, outdoor terrain, hospitals. Ready in minutes, not months.

500+ environments

Task Libraries

Pre-configured task definitions with reward functions, success criteria, and observation spaces. Don't write reward code. Just train.

pick, place, grasp, fold...

Training Curricula

Ordered difficulty progressions proven to produce generalizable policies. Start easy, end expert. Built by researchers who've done it before.

progressive difficulty

Domain Randomization

10,000 object variations per environment. Tested randomization configs that maximize transfer rate without breaking policy learning.

sim-to-real optimized

Genesis-Powered Physics

Built on Genesis — the physics engine that runs 430,000x faster than real time. Train policies in 26 seconds that used to take days.

430,000x speedup

Version Control for Envs

Track environment versions, compare policy performance across variants, roll back to previous configs. Git for simulation.

environment versioning

Community Contributions

Share environments with the Kyros community. Earn credits when others use your work. The more the community builds, the better everyone trains.

open marketplace

Transfer Validation

Every Kyros environment is validated against real-hardware transfer benchmarks. Verified environments carry a 'Transfer-Tested' badge.

transfer-verified

See the training worlds we've built.

Kitchen_residentialActive
variations: 10,247transfer_rate: 94.2%
Warehouse_fulfillmentActive
variations: 10,247transfer_rate: 94.2%
Outdoor_terrainActive
variations: 10,247transfer_rate: 94.2%
Manufacturing_lineActive
variations: 10,247transfer_rate: 94.2%
Medical_roomActive
variations: 10,247transfer_rate: 94.2%
Validation

What researchers say about the problem we're solving.

"
The sim-to-real gap remains a major obstacle. Agents trained in simulators often struggle to maintain performance when transferred to real-world physical environments.
— Published Research, IEEE Robotics Survey, 2024
sim-to-real gap
"
Creating realistic physical simulations requires extensive technical expertise, expensive hardware, and time-consuming manual processes. Existing tools often fail to deliver the speed, accuracy, and user-friendliness needed.
— Robotics & Embodied AI Research Survey, 2024
tooling gap
"
The robotics and embodied AI field has long struggled with accessibility and efficiency issues — making robotics research an exclusive domain for well-funded institutions.
— Physical AI Infrastructure Analysis, 2024
access gap

The robots of tomorrow need training grounds today.

Join the waitlist. Be the first team to access Kyros.