Three main ways to generate it
Simulation generates data by modeling a physical or logical environment and recording what happens inside it. Self-driving car development uses driving simulators that render realistic roads, weather, and traffic, letting engineers generate sensor data for scenarios that would be dangerous or impractical to stage with a real vehicle, like a child running into the street or a sudden tire blowout at highway speed.
Procedural generation uses rules and algorithms to construct data programmatically rather than simulating physics. This might mean generating synthetic documents by combining templates with randomized fields, or building varied 3D scenes from a library of objects and layout rules. It's especially common for tasks where the underlying structure is well understood but real examples are scarce.