Technical GlossaryData Science and Data Management

Domain Randomization

An approach that varies environmental factors in synthetic data generation to make models more robust to the real world.

Domain randomization is a powerful approach used especially in sim-to-real transfer problems. Lighting, color, background, viewpoint, noise, and physical parameters are systematically varied. The goal is to prevent the model from overfitting to a single clean synthetic distribution and make it more robust. This approach has delivered important results particularly in robotics and computer vision. The central idea is not to replicate reality perfectly, but to represent a broad enough range of plausible variation.