Synthetic data helps scale robot training by generating diverse, controllable scenarios, reducing physical data collection demands, enabling edge-case training, and improving embodied AI performance through simulation and real-world data.