Emergence of Norms in Interactions with Complex Rewards

Autonomous systems are becoming pervasive, and as they become applied to highly dynamic and heterogeneous environments there is a need to model and understand more complex and nuanced agent interactions than have been previously studied. This paper proposes an agent-based modelling approach, based on norm emergence, to investigate such interactions. While there is typically an ideal set of compatible actions which lead to an optimal norm, in complex environments there may also be combinations that are compatible and yield positive (but not optimal) rewards. We illustrate our model of such scenarios using the case of an autonomous vehicle performing a manoeuvre at a T-intersection.