A knowledge networking approach for AI-driven roundabout risk assessment

Smart applications in vehicular networks, such as highly-automated driving, require knowledge to support complex decision making which is highly dependent on the current driving context, for example, through machine learning based object recognition. Unlike information, the pertinence of a knowledge model depends on its context of use, rather than its date of creation. In turn, the existing information sharing mechanisms in vehicular networks, optimized for fast information delivery, must be adapted to support rich contextual queries, and let vehicles discover the right knowledge for the right context. Moreover, networking of knowledge models has the potential to alleviate the redundant transmission and computation of similar information. Through a case of vehicle exit probability knowledge distribution in a roundabout, we show the impact and potential of a context-based dissemination of knowledge in terms of accuracy, delay, and overhead compared to context-agnostic approaches.