An Affective Anticipatory Agent Architecture

Abstract

We present an agent architecture for an enhanced theory of intent prediction with affective evaluation of expectations. The architecture combines models from psychology and robotics to create an online situation appraisal mechanism for preferential evaluation of predicted future states, which determines action selection in the current state. The architecture is implemented in a vehicular agent situated in a motorway driving scenario, requiring more than goal-directed planning: the agents must model the behaviour of others, and predict and evaluate future states. Simulations are described, showing that global properties emerge in the system that are improved and more stable with the new agent architecture.