Assisted and Automated Driving (AAD) Functions are inherently limited by the definition of their Operational Design Domain (ODD) from one side, whilst from the other side the “definition” of Operational Design Domain remain incomplete and linked to some attributes of the environment, road infrastructure, etc., but does not accurately cover for real-world variability. For example, a snow rate cannot accurately capture the complexity of different snow phenomena, there are no established ways to define road or snow/spray, etc.
This tutorial aims to challenge this limit, involving the participants to reflect and discuss on the following topics, to collectively tackle and align understanding on how to use “ODD” in a more adaptive (and safer) way.
– What are the external and internal factors that affect the capacity of the ADD system to build a complete situational awareness?
– Which factors are important to define within the ODD, and do we need an expanded definition?
– Are we able to re-use existing information on vehicle to classify with adequate accuracy the needed “ODD” attributes?
– What do AAD systems need to detect to predict proactively variations of the ODD?
– Are there emerging sensing technologies that can enhance current capabilities?
– Can we use real time perception sensors’ quality to adapt to the ODD dynamically?
Given that data is at the basis of the autonomy stack, this tutorial aims to revolutionise current perspective on ODD, proposing that, in a ‘human like fashion’, ADD functions should use data quality, redundancy, and cross checks as the way to establish during operations when the system can work safely and how to ‘adapt’ based on the dynamically detected ODD.
