Bio-inspired decision making system for an autonomous social robot: the role of fear

• Autores: Álvaro Castro González
• Directores de la Tesis: María de los Ángeles Malfaz Vázquez (dir. tes.), Miguel Ángel Salichs Sánchez-Caballero (dir. tes.)
• Lectura: En la Universidad Carlos III de Madrid ( España ) en 2012
• Idioma: inglés
• Tribunal Calificador de la Tesis: Carlos Balaguer Bernaldo de Quirós (presid.), Antoni Gomila Benejam (secret.), Fernando Silva (voc.)
• Enlaces
  • Tesis en acceso abierto en: e-Archivo
• Resumen

  • Robotics is an emergent field which is currently in vogue. In the near future, many researchers anticipate the spread of robots coexisting with humans in the real world. This requires a considerable level of autonomy in robots. Moreover, in order to provide a proper interaction between robots and humans without technical knowledge, these robots must behave according to the social and cultural norms. This results in social robots with cognitive capabilities inspired by biological organisms such as humans or animals. The work presented in this dissertation tries to extend the autonomy of a social robot by implementing a biologically inspired decision making system which allows the robot to make its own decisions. Considering this kind of decision making system, the robot will not be considered as a slave any more, but as a partner. The decisionmaking systemis based on drives,motivations, emotions, and self-learning. According to psychological theories, drives are deficits of internal variables or needs (e.g. energy) and the urge to correct these deficits are the motivations (e.g. survival). Following a homeostatic approach, the goal of the robot is to satisfy its drives maintaining its necessities within an acceptable range, i.e. to keep the robot’s wellbeing as high as possible. The learning process provides the robot with the proper behaviors to cope with each motivation in order to achieve the goal. In this dissertation, emotions are individually treated following a functional approach. This means that, considering some of the different functions of emotions in animals or humans, each artificial emotion plays a different role. Happiness and sadness are employed during learning as the reward or punishment respectively, so they evaluate the performance of the robot. On the other hand, fear plays a motivational role, that is, it is considered as a motivation which impels the robot to avoid dangerous situations. The benefits of these emotions in a real robot are detailed and empirically tested. The robot decides its future actions based on what it has learned from previous experiences. Although the current context of this robot is limited to a laboratory, the social robot cohabits with humans in a potentially non-deterministic environment. The robot is endowed with a repertory of actions but, initially, it does not know what action to execute either when to do it. Actually, it has to learn the policy of behavior, i.e. what action to execute in different world configuration, that is, in every state, in order to satisfy the drive related to the highest motivation. Since the robot will be learning in a real environment interacting with several objects, it is desired to achieve the policy of behavior in an acceptable range of time. The learning process is performed using a variation of the well-known Q-Learning algorithm, the Object Q-Learning. By using this algorithm, the robot learns the value of every state-action pair through its interaction with the environment. This means, it learns the value that every action has in every possible state; the higher the value, the better the action is in that state. At the beginning of the learning process these values, called the Q values, can all be set to the same value, or some of them can be fixed to another value. In the first case, this implies that the robot will learn from scratch; in the second case, the robot has some kind of previous information about the action selection. These values are updated during the learning process. The emotion of fear is particularly studied. The generation process of this emotion (the appraisal) and the reactions to fear are really useful to endow the robot with an adaptive reliable mechanism of “survival”. This dissertation presents a social robot which benefits from a particular learning process of new releasers of fear, i.e. the capacity to identify new dangerous situations. In addition, by means of the decision making system, the robot learns different reactions to prevent danger according to different unpredictable events. In fact, these reactions to fear are quite similar to the fear reactions observed in nature. Another challenge is to design a solution for the decision making system in such a way that it is flexible enough to easily change the configuration or even apply it to different robots. Considering the bio-inspiration of this work, this research (and other related works) was born as a try to better understand the brain processes. It is the author’s hope that it sheds some light in the study of mental processes, in particular those which may lead to mental or cognitive disorders. -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------