Demonstrators

SCD 7.1: Enriched virtual models based on standardized real-world data (lead: AVL)

AVL

AIT, TUGRAZ

With a focus on reliable and robust ADAS/AD systems, demonstrator SCD7.1 advances this vision by developing sophisticated methodologies for intelligent data collection, automated virtualization, and AI controller learning. This involves introducing dedicated data collection and processing pipelines to capture real-world data and virtualize it using standardized formats. These virtualized scenarios can then be utilized for AI-based controller learning strategies, as well as for validation and evaluation purposes through advanced monitoring systems. This comprehensive approach ensures the development of reliable and effective ADAS/AD systems.

With the goal of designing demonstrator SCD7.1 so that all developed components function both within a complete toolchain and as standalone tools, numerous deployment and utilization possibilities arise. Data collection and virtualization can be utilized to investigate and evaluate the performance of ADAS/AD systems or potential misbehaviors in a safe environment. Furthermore, virtualized scenarios can be used for training AI driving controllers, where the second part of the demonstrator plays an important role. The learning framework enables the training of controllers based on a diverse set of virtualized scenarios, allowing an agent to interact with the environment and receive feedback. The validation and evaluation framework, responsible for validating ADAS/AD systems using implemented safety monitors, enables identifying potential problems in the system under test and provides valuable feedback for ADAS/AD development.

Figure 1: High-level system level desing of demonstrator SCD7.1.