Quality, variety and ODD coverage of sensor data and annotations are crucial to develop and test perception and fusion modules. Collecting data in the real world only partly fulfill these requirements, hence synthetic sensor data can help to fill the gaps. Especially when a specific coverage hole needs to be filled quickly, real-world data is not a solution. Synthetic sensor data in various levels of detail is promising help, but currently often fails in real-world use due to complex simulation setup and scene modeling, inflexible annotations and inefficient data generation processes. In this presentation, we introduce and discuss new tools for efficient perception and fusion development and testing tailored to make synthetic sensor data accessible and helpful.

In this session, you will learn:

• How synthetic sensor data can help training and testing perception modules
• Pros and cons of generative AI and simulative methods
• How to configure annotations according to your needs
• Why annotations are important during end-to-end testing