Tesla and Waymo use synthetic data to train perception modules

Companies such as Tesla and Waymo are using synthetic data generated by Diffusion models to train their perception modules. These synthetic data can provide a variety of complex driving scenarios, such as extreme weather and sudden obstacles, to help the model generalize better. In addition, these data can also simulate high-fidelity LiDAR point clouds, radar data, or images under different lighting conditions to make up for the lack of real data. In this way, these companies can reduce their reliance on labeled data and further improve the performance of the model.