A Semantic Model for Information Sharing in Autonomous Vehicle Systems

Abstract

Autonomous driving has gained increasing attention and achieved significant successes. To enable autonomous driving, a vehicle needs to collect sensor data and process them in real-time to derive a perception model of the surrounding environment. Driving control decisions are derived based on the perception model. However, the semantics of the perception model are relatively primitive. Some additional semantic information can be helpful in autonomous driving decision making. In this paper, we build a rich semantic model to capture the relevant semantic information in autonomous vehicle systems. The model can help improve autonomous driving in several aspects. First, a well-defined semantic model can facilitate information exchange and integration across multiple vehicles, which can help build an extended view of the environment and improve driving decisions. Second, contextual information such as weather, time, calendar events, etc., can help understand the potential traffic condition of the environment and enhance the predictions of the possible motions of mobile objects in the environment. Third, to support learning from existing driving experiences, some semantic information can be used to help associate relevant situations and facilitate effective learning. We further discuss how to use our semantic model to enhance autonomous driving.