Simulation

No one is going to build genuinely intelligent robots purely in the physical world. It’s too slow, too expensive, and too dangerous.

Building intelligence for robots requires more than just algorithms. It requires experience - thousands of hours of interaction with the world that would be impractical or dangerous in physical environments.

Intelligence requires experience. Experience requires simulation.

We are building a (virtual) robotics R&D lab, to lower the cost of experimentation and failure in robotics.

• Physics-based simulation environment for robotics
• Realistic digital twins of popular robots (e.g. Unitree, Engine01, etc), down to the APIs
• Build and train your robotic AI models in a realistic simulation environment
• Provision high-quality, realistic scenarios - e.g. warehouse, home
• 1000x cheaper than working on an actual robots

When robots trained in our simulation hit the real world, they behave as expected - not as hoped.

Digital twins

We’ve created precise digital twins of popular robots:

• Unitree quadrupeds with exact joint limits and dynamics
• Engine01 humanoids with accurate mass distribution and balance characteristics
• Manipulator arms with proper payload constraints and precision modeling
• Identical API interfaces to their physical counterparts

This means zero translation between simulation and reality.

Realistic environments

Our simulation provides high-quality scenarios:

• Warehouses with varying inventory and obstacles
• Homes with furniture, objects, and human interaction
• Offices with dynamic elements and navigation challenges
• Outdoor environments with terrain and weather variation

1,000x cheaper than hardware

Physical robots are prohibitively expensive:

• $15,000+ for basic quadrupeds
• $50,000+ for capable humanoids
• Thousands in inevitable repair costs

Our simulation gives you unlimited virtual robots at a fraction of the cost.

1,000x faster iteration

Simulation accelerates development by orders of magnitude:

• Run thousands of scenarios in parallel
• Test a year’s worth of operations in a day
• Deploy to physical robots only when ready

Browser-based access

Our simulation runs in your browser.

BUILT FOR AI DEVELOPMENT Our simulation is a complete development environment:

• Train reinforcement learning models in-browser
• Integrate with LLMs for planning and reasoning
• Capture demonstration data for imitation learning
• Test perception systems with realistic sensor inputs
• Deploy models to physical robots with one click

Why?

Every transformative technology follows the same pattern: simulate extensively, then build physically.

• Robotics has traditionally moved slowly due to the high cost of failure
• We want to lower the cost of experimentation and failure in robotics
• Our strategy is to focus on cheap progress, and then solve for Sim2Real as an expensive subsequent problem