Preference-Oriented Planning for Achieving Satisfactory Emotional States

During the last years the study of emotions and the characteristics of the human personality is increasingly relevant. This work proposes a deliberative emotional model for virtual agents based on their basic needs and preferences, implemented as an automated planning domain. The domain describes the knowledge of one agent living in a virtual world that has a set of basic physiological needs as feeding, hydration, rest, hygiene and entertainment. To cater for its needs, it must carry out different activities interacting with different objects over which it has preferences of one object over another. The selection depends on its preferences and directly affects its emotional state. So, the plans generated by the system attempt to fulfil the agent’s physiological needs, choosing the objects required by each action to achieve the most satisfactory emotional state. Introduction Emotions have been shown in other fields to play an important role on rationality. Thus, the work on understanding and using emotions has been extended to the Artificial Intelligence field. Among other things, emotions can explain part of the interaction of the individual with the environment. This allows rational agents to approach all those objects and agents in the environment that cater for its needs and benefit its well-being. And, also, it allows those agents to move away of objects that endanger its survival (Breazeal 2003). Emotions have been studied in Psychology, Neurology and Physiology from a wide variety of points of view and each field focuses the attention on different aspects. Despite everything, it seems there is some agreement to consider emotions as an inborn and subjective reaction to the environment, with an adaptative function, and accompanied of several organic, physiological and endocrine changes. Another point of agreement is that emotions are an outstanding factor in humans and living beings, because they modify the usual behaviour depending on changes in the environment. In the development of systems that interact with persons, as human behaviour simulators, emotions should not be ignored, because, on one hand, they may help on this interaction and, on the other hand, they constitute a decisive part of human reasoning and behaviour. During the last years, several emotion-oriented systems have been developed, that normally follow Frijda’s theory about emotions based on a functional hypothesis: emotions are functional most of the time (Frijda 1995). Thus, the use of emotions in artificial systems is needed to achieve an objective. As an example, Cañamero proposes a homeostatic approach to the motivations model. She creates a selfregulatory system, very close to natural homeostasis, that connects each motivation to a physiological variable, which is controlled within a given range (Cañamero 1997). When the value of that variable differs from the ideal one, an error signal proportional to the deviation, called drive, is sent, and activates some control mechanism that adjusts the value to the right direction. There are other architectures based on drives, as the Dorner’s PSI architecture used by Bach (Bach & Vuine 2003) and also by Lim (Lim et al. 2005), that offer a set of drives of different type, as certainty, competence or affiliation. Taking as reference Cañamero’s and Gadanho’s works on how to improve artificial emotions in robot’s behaviour (Gadanho & Hallam 2001), Malfaz created a decision-making system based on emotions and machinelearning (through reinforcement learning) for social autonomous agents (Malfaz & Salichs 2006). Her agents developed artificial emotions, like happiness and fear, and learned a way of behaving that let them avoid some risks and keep a standard well-being. However, most works on emotional agents are based on reactive behaviours, so there is no inference being done on medium-long term goals and the influence of emotions on how to achieve those goals. There are some works on emotions based on planning, but mainly oriented to storytelling. Examples are emergent narrative in FEARNOT! (Aylett et al. 2005) and the interactive storytelling of Madame Bovary on the Holodeck (Cavazza et al. 2007). In the present work, a model of long term reasoning based on emotions has been designed following some ideas introduced by Avradinis and colleagues (Avradinis et al. 2003) using some concepts that already appear in Cañamero’s and Malfaz’s works, like motivations and the use of drives to represent basic needs. The model has been implemented as an automated planning domain that constitutes the reasoning core of a possible client in the virtual and multi-agent world AI-LIVE (Fernández et al. 2008). AI-LIVE is a client/server application oriented towards the intensive use of Artificial Intelligence controlled Bots, and it was designed as a test environment of several Artificial Intelligence techniques. It borrows the idea from the popular video game THE SIMS, where the player creates individual characters (units) that have significant autonomy, with their own drives, goals, and strategies for satisfying those goals. In this implementation, we introduce the concept of how an agent prefers some objects and the influence of those on long term achievement of goals. The paper presents first the model design. Then, we describe the domain that implements the model, and show some empirical results that validate the model. Finally, we draw some conclusions derived from the work, and propose future research lines.