Learning social codes for self-interested agents

Abstract

This thesis will explore interaction between non-supervised machinelearning agents, with a focus on traffic situations. We intend to showthat self-interested agents can arrive at mutually beneficial strategiesaccording to general game theory, achieving pure or mixed Nash equi-libria, and that this has practical applications for unmanned vehicles.We will start by verifying against simple one stage models, specifi-cally “battle of the sexes” and “chicken race”, before extending this touncontrolled four way intersections.