The development of perception systems for autonomous vehicles requires high volumes of highly accurate and domain-specific data to train and validate algorithms effectively. Using synthetic data allows perception system performance to be unlocked from the limitations of relying on real-world data alone – which is expensive, time-consuming to collect and annotate and difficult to scale to the required volumes. The presentation will highlight the development of advanced simulation camera models with tuneable CMOS and ISP features, which are capable of rapidly expanding the training datasets for perception systems. It will also explore the methodologies employed to validate the models and provide insights gained from applying synthetic data to real-world systems.