Synthetic data

Data created through simulation, procedural generation or generative models rather than directly recorded from the target setting.

Synthetic examples can expand coverage and provide known scene attributes. They still need checks for physical plausibility, relevance and differences from real conditions. Accurate labels inside a simulator do not establish that a model trained on them will perform reliably in deployment.

In practice

A simulated grasp dataset varies lighting and object pose, then is evaluated for relevance to the robot's real operating conditions.