Definition
Domain randomization trains policies in simulation with heavily randomized visual appearance (textures, lighting, colors, camera positions) and physical parameters (friction, mass, damping). The intuition is that if a policy succeeds across a wide distribution of simulated environments, the real world becomes just another sample from that distribution. Visual domain randomization has enabled direct sim-to-real transfer for grasping and locomotion. Physics randomization helps bridge the reality gap for contact-rich tasks. It is a core technique in sim-to-real transfer pipelines alongside system identification and domain adaptation.
Why It Matters for Robot Teams
Understanding domain randomization is essential for teams building real-world robot systems. Whether you are collecting demonstration data, training policies in simulation, or deploying in production, this concept directly affects your workflow and system design.