This presentation explores a data-driven approach to developing and validating perception functions by harmonizing synthetic and real-world datasets. It depicts a workflow starting with real-world data collection, followed by reproduction in HIL environments. Missing edge cases are addressed by introducing synthetic sensor and bus data, utilizing AI-driven enhancement methods (e.g. NVIDIA Cosmos) to reduce the reality gap. Finally, it demonstrates the transition to end-to-end testing, using identical real-time systems for both open- and closed-loop validation. This mixed-reality approach addresses the challenges of real-world testing and enhances the robustness of perception function testing by utilizing cutting-edge simulation environments and AI.

• How to use real-world data to test perception functions
• How to generate synthetic datasets of rare edge case scenarios
• How to reduce the reality gap of synthetic data
• How to use a modular real-time environment for open- and closed-loop testing