26 Oct 2026 · Vision
Why every robot will be born in simulation
Physical AI cannot afford to learn from its mistakes in the real world. Here is how we think about getting robots from simulation to deployment.
Digital AI mostly lives in a virtual world. When it gets something wrong, the damage is usually small and a fix is a deploy away. Physical AI does not get that luxury. A robot that makes a mistake on an assembly line, in a warehouse or in a hospital can break equipment worth more than the robot, and it can hurt people.
So the question for anyone deploying robots is not only can it do the task, but how sure are we that it will.
Simulation is where that certainty comes from
We believe that in the future, every robot will need a simulation of the place where it is going to work. That simulation is where it is trained, where it is tested, and where it has to clear a measured bar of reliability before anyone lets it loose in the real world.
That only works if simulation is faithful. A policy that learns on objects with the wrong mass, colliders that do not match the surface, or a generic room instead of the actual factory cell will learn the wrong lessons. The loss between what works in simulation and what works in reality is the sim-to-real gap, and closing it is the whole game.
Real to sim to real
We think about the problem as one pipeline:
- R2S, real to sim: bring the real world, the objects and the spaces a robot will work in, into simulation.
- S2R, sim to real: make that simulation behave like reality, so policies trained and tested there transfer back.
Piece by piece
The full problem is too big to solve in one step, so we are solving it in pieces, each useful on its own.
We started with the smallest unit every simulation is made of: the asset. Our pipeline turns real objects, or plain-language descriptions of them, into SimReady assets, with geometry, colliders, mass, inertia, friction, materials and joints, autonomously. Rigid and articulated objects work today. Deformables are next.
From there we move to whole scenes, then to reconstructing real deployment sites, then to evaluations and closing the loop with real-world feedback.
Every robot is going to be born in a simulation. We want to build the place it is born.
If you are putting robots to work in the real world, we would like to hear from you.