Action selection through affective states modelling

We introduce an action selection framework for the advanced behavioural animation of virtual creatures. In modern creative media, the behavioural animation of characters which act in a believable fashion is an ongoing challenge. Traditional action selection approaches which attempt to make an agent act rationally often fall short of the believability required for the modern consumer. Often the most believable action is not the most rational one, and our judgement of an agent's behaviour may also be based on the perception of its personality. Our approach, Affective Spaces Modelling, addresses these issues by creating a multi-dimensional environment constructed of aspect dimensions, with each aspect dimension representing a linear scale of a single component of the agent's internal state. Affective states can then be modelled by placing them in a single point in this environment. As the agent's state changes within the affective state space, different affects trigger appropriate actions. We demonstrate through a case study how the technique can be used to simulate different types of agent behaviour, operating both individually and as part of a group. Our case studies focus on groups of agents, allowing for the direct comparison of different personalities and examples of behavioural phenomena.