With the increasing complexity of autonomous functionality in both AV and ADAS systems, traditional methods for developing safety-critical software are becoming inadequate. Since autonomous systems need to operate in complicated real-world driving domains, they are expected to handle a near endless variety of scenarios. This means system requirements have to be added and adjusted as new edge cases and scenarios are encountered throughout the vehicle development process.

‍

‍

Deontic proposes an automated software solution to augment and extend such driving scenarios. Our solution finetunes LLMs (Large Language Models) on scenario description languages like OpenSCENARIO to include abstract knowledge represented in many safety standards, laws and regulations.

‍

In most cases, this knowledge is represented in an abstract form like text, which cannot be directly used in a technical process. In order to use this information in technical processes, reviews or technical preparations are necessary. For example, standards and guidelines are used to define minimal and maximal allowed values for different scenario parameters. Furthermore, these sources of information are used to create a common understanding of the operational design domain (ODD). This is important especially if the test object should be released in different countries. The abstract knowledge is therefore a valuable source for describing and defining AD scenarios.