All terms

Glossary · AI & ML

Domain randomization

In brief

Domain randomization is the technique of training a policy across many randomized variants of the simulator — different physics parameters, lighting, textures, sensor noise — so the policy learns to be robust to the variation. It is the standard tool for closing the sim-to-real gap.

A policy trained in a single perfectly-tuned simulator overfits to that simulator. Run the same policy on a real robot and it breaks: real friction is different, real latency is different, real sensor noise has a different distribution. Domain randomization defends against this by varying the simulator parameters during training and forcing the policy to generalize.

How much randomization is enough is project-specific. Too little and sim-to-real fails; too much and the policy becomes overly conservative or just doesn't learn. The art is in picking the variation distributions to match real-world uncertainty.

Related terms

See it in the wild

Browse robots and brands using these techniques

Glossary entries are upstream. The catalog is where the implementations live.