ThesisPDF Available

Towards Emotional Modulation of Action Selection in Motivated Autonomous Robots


Abstract and Figures

“Behavior-based” artificial intelligence (AI) proposes a “bottom-up” or “synthetic” approach to study intelligence in the context of “complete” autonomous agents. This paradigm is highly inspired by animal behavior, neuroscience and evolution theory, and it is appropriate to deal with situations that require a system to autonomously carry out several time-dependent tasks in a dynamic, unpredictable environment. One of the main problems arising in those situations is action selection, i.e., making a decision as to what behavior to execute next in order to carry out several conflicting tasks and survive in the environment. Although many action selection architectures have been proposed, the literature lacks systematic ways to analyze their properties and their performance. In this thesis I firstly propose both quantitative and qualitative analyses of action selection architectures. I compare different motivation-based architectures within virtual and robotic frameworks and show the adequacy of “winner-take-all” and “voting-based” architectures for different environmental conditions. I then show that these architectures can greatly improve their adaptivity if modulated by second-order controllers. Inspired by neuroscience, I propose underlying mechanisms of emotions as second-order controllers or modulators of motivation-based action selection architectures. Acting only at the sensory level of the architectures and without modifying their structure, these mechanisms generate behavior adapted to changing environmental conditions. As a result of such modulation different functionalities emerge. This research departs from more “top-down” engineering approaches to emotion modeling, which traditionally implement those functionalities as predefined “emotion modules”.
Content may be subject to copyright.
A preview of the PDF is not available
... The specific implementation of the cue-deficit model in the case of [8] used a simple product of bounded drive values (deficits of essential variables) but in principle could be adapted to more complex functions of cues and deficits. [34,35], for example, who explicitly used Ashby's notion of essential variables, implemented a variation on McFarland and Spier's ( [8]) original cue-deficit model. In their work 'deficits' in essential variables, e.g. ...
Conference Paper
Full-text available
In this chapter, we present a minimalist approach to utilizing the computational principles of affective processes and emotions for autonomous robotics applications. The focus of this paper is on the presentation of this framework in reference to preservation of agent autonomy across levels of cognitive-affective competences. This approach views autonomy in reference to (i) embodied (e.g. homeostatic), and (ii) dynamic (e.g. neural-dynamic) processes, required to render adaptive such cognitive-affective competences. We hereby focus on bridging bottom-up (standard autonomous robotics) and top-down (psychology-based dimensional theoretic) modelling approaches. Our enactive approach we characterize according to bi-directional grounding (inter-dependent bottom-up and top-down regulation). As such, from an emotions theory perspective, ‘enaction’ is best understood as an embodied and dynamic appraisal perspective. We attempt to clarify our approach with relevant case studies and comparison to other existing approaches in the modelling literature.
... This follows the thinking of McFarland and Spier (1997; McFarland and Bösser, 1993; McFarland, 2008) who have advocated the need for robots to react to real-time environmental opportunities (opportunism) when considering homeostatic regulation of multiple internal needs and goal/need directives. Garcìa (2004), Avila-Garcìa and Cañamero (2005) have suggested that emotion-relevant synthetic hormones allow robotic agents to trade off the need for opportunism with the need to persist in the pursuit of a particular need-fulfilling goal. More recently, work by Lowe et al. (2010a), Montebelli et al. (2010 has considered the importance of different types of energy constraints to the homeostatic regulation of adaptive behavior of robotic agents with multiple goals. ...
Full-text available
In this article, we review the nature of the functional and causal relationship between neurophysiologically/psychologically generated states of emotional feeling and action tendencies and extrapolate a novel perspective. Emotion theory, over the past century and beyond, has tended to regard feeling and action tendency as independent phenomena: attempts to outline the functional and causal relationship that exists between them have been framed therein. Classically, such relationships have been viewed as unidirectional, but an argument for bidirectionality rooted in a dynamic systems perspective has gained strength in recent years whereby the feeling-action tendency relationship is viewed as a composite whole. On the basis of our review of somatic-visceral theories of feelings, we argue that feelings are grounded upon neural-dynamic representations (elevated and stable activation patterns) of action tendency. Such representations amount to predictions updated by cognitive and bodily feedback. Specifically, we view emotional feelings as minimalist predictions of the action tendency (what the agent is physiologically and cognitively primed to do) in a given situation. The essence of this point is captured by our exposition of action tendency prediction-feedback loops which we consider, above all, in the context of emotion regulation, and in particular, of emotional regulation of goal-directed behavior. The perspective outlined may be of use to emotion theorists, computational modelers, and roboticists.
... We use affect to describe the exogenous component of comfort as it is described in (Likhachev andArkin 2000, Dunn 1977) and in Chap. 5 we make precise how we use and compute this definition in robotics. The endogenous component of comfort is "well-being" which represents satisfaction of the primary needs (Avila-García 2006, Likhachev and Arkin 2000, Kahneman, Diener and Schwarz 1999, Dunn 1977 and we make precise our robotics use of this notion in Chap. 4. We therefore present the working definition of well-being. ...
In animals, humans and robots, imitative behaviours are very useful for acting, learning and communicating. Implementing imitation in autonomous robots is still a challenge and one of the main problems is to make them choose when and who to imitate. We start from minimalist architectures, following a bottom-up approach, to progressively complete them. Based on imitation processes in nature, many architectures have been developed and implemented to increase quality (essentially in terms of reproducing actions with accuracy) of imitation in robots. Nevertheless, autonomous robots need architectures where imitative behaviour is well integrated with the other behaviours like seeking for stability, exploration or exploitation. Moreover, whether to express imitative behaviours or not should also depend on the history of interactions (positive or negative) between robots and their interactive partners. In this thesis, we show with real robots how low-level imitation can emerge from other essential behaviours and how affect can modulate the way they are exhibited. On top of proposing a novel vision of imitation, we show how agents can autonomously switch between these behaviours depending on affective bonds they have developed. Moreover, with simple architectures, we are able to reproduce behaviours observed in nature, and we present a new way to tackle the issue of learning at different time scales in continuous time and space with discretization.
The somatic marker hypothesis (SMH) posits that the role of emotions and mental states in decision-making manifests through bodily responses to stimuli of import to the organism's welfare. The Iowa Gambling Task (IGT), proposed by Bechara and Damasio in the mid-1990s, has provided the major source of empirical validation to the role of somatic markers in the service of flexible and cost-effective decision-making in humans. In recent years the IGT has been the subject of much criticism concerning: (1) whether measures of somatic markers reveal that they are important for decision-making as opposed to behaviour preparation; (2) the underlying neural substrate posited as critical to decision-making of the type relevant to the task; and (3) aspects of the methodological approach used, particularly on the canonical version of the task. In this paper, a cognitive robotics methodology is proposed to explore a dynamical systems approach as it applies to the neural computation of reward-based learning and issues concerning embodiment. This approach is particularly relevant in light of a strongly emerging alternative hypothesis to the SMH, the reversal learning hypothesis, which links, behaviourally and neurocomputationally, a number of more or less complex reward-based decision-making tasks, including the ‘A-not-B’ task – already subject to dynamical systems investigations with a focus on neural activation dynamics. It is also suggested that the cognitive robotics methodology may be used to extend systematically the IGT benchmark to more naturalised, but nevertheless controlled, settings that might better explore the extent to which the SMH, and somatic states per se, impact on complex decision-making.
The fact that emotions are considered to be essential to human reasoning suggests that they might play an important role in autonomous robots as well. In particular, the decision of when to interrupt ongoing behavior is often associated with emotions in natural systems. The question under examination here is whether this role of emotions can be useful for a robot which adapts to its environment. For this purpose, an emotion model was developed and integrated in a reinforcement-learning framework. Robot experiments were done to test an emotion-dependent mechanism for the automatic detection of the relevant events of a learning task against more traditional approaches. Experimental results are presented that confirm that emotions can be useful in this role, specifically by improving the efficiency of the learning algorithm.
Network models have started to be used to control physical robots. A major problem is how to control the selection of grossly different high level behaviors and how to express that selection down at the motor level. In this paper we show how to extend the subsumption architecture [Brooks 86] inspired by a model of behavior selection of [Maes 89], and incorporating a model of hormonal control [Kravitz 88]. In robotic terms a hormonal control system enables global control of a distributed system with only a very low bandwidth global communication system. This has important implications concerning the ability to fabricate intelligent silicon controllers for tiny microfabricated micro robots [Flynn, Brooks and Tavrow 89].