Development and Implementation of a Pipeline for the Simulative Generation of Training Data for Lidar Object Detectors

This thesis presents a simulation-based pipeline for generating synthetic training data for LiDAR object detectors in autonomous vehicles. By integrating procedural content generation (PCG) with simulation frameworks, the pipeline automates the creation of diverse, high-quality training datasets. It employs the Wave Function Collapse (WFC) algorithm for environment generation, ASAM Open-DRIVE/OpenSCENARIO standards for scenario modelling, a Functional Mock-up Unit (FMU) for LiDAR simulation via the Functional Mock-up Interface (FMI), and OSI for standardized sensor data exchange and interoperability. Automated labelling ensures accurate annotations, reducing the need for manual effort. The approach enhances dataset diversity, improves scalability, and ensures compatibility with existing development frameworks, offering a cost-effective alternative to real-world data collection.