Ironies of social robotics.

Aiming for "humanlike" or "natural" interactions can make social robots and their limitations more difficult to understand.

How puzzling is the social artifact puzzle?

In this commentary we would like to question (a) Clark and Fischer's characterization of the "social artifact puzzle" - which we consider less puzzling than the authors, and (b) their account of social robots as depictions involving three physical scenes - which to us seems unnecessarily complex. We contrast the authors' model with a more parsimonious account based on attributions.

Understanding Social Robots: Attribution of Intentional Agency to Artificial and Biological Bodies.

Much research in robotic artificial intelligence (AI) and Artificial Life has focused on autonomous agents as an embodied and situated approach to AI. Such systems are commonly viewed as overcoming many of the philosophical problems associated with traditional computationalist AI and cognitive science, such as the grounding problem (Harnad) or the lack of intentionality (Searle), because they have the physical and sensorimotor grounding that traditional AI was argued to lack. Robot lawn mowers and self-driving cars, for example, more or less reliably avoid obstacles, approach charging stations, and so on-and therefore might be considered to have some form of artificial intentionality or intentional directedness. It should be noted, though, that the fact that robots share physical environments with people does not necessarily mean that they are situated in the same perceptual and social world as humans. For people encountering socially interactive systems, such as social robots or automated vehicles, this poses the nontrivial challenge to interpret them as intentional agents to understand and anticipate their behavior but also to keep in mind that the intentionality of artificial bodies is fundamentally different from their natural counterparts. This requires, on one hand, a "suspension of disbelief " but, on the other hand, also a capacity for the "suspension of belief." This dual nature of (attributed) artificial intentionality has been addressed only rather superficially in embodied AI and social robotics research. It is therefore argued that Bourgine and Varela's notion of Artificial Life as the practice of autonomous systems needs to be complemented with a practice of socially interactive autonomous systems, guided by a better understanding of the differences between artificial and biological bodies and their implications in the context of social interactions between people and technology.

Do we really want AI to be human-like?

Behavioral variability can be used to make robots more human-like, but we propose that it may be wiser to make them less so.

Understanding robots.

Elucidating the neural and psychological mechanisms underlying people's interpretation of robot behavior can inform the design of interactive autonomous systems, such as social robots and automated vehicles.

Do You See what I See? Tracking the Perceptual Beliefs of Robots.

Keeping track of others' perceptual beliefs-what they perceive and know about the current situation-is imperative in many social contexts. In a series of experiments, we set out to investigate people's ability to keep track of what robots know or believe about objects and events in the environment. To this end, we subjected 155 experimental participants to an anticipatory-looking false-belief task where they had to reason about a robot's perceptual capability in order to predict its behavior. We conclude that (1) it is difficult for people to track the perceptual beliefs of a robot whose perceptual capability potentially differs significantly from human perception, (2) people can gradually "tune in" to the unique perceptual capabilities of a robot over time by observing it interact with the environment, and (3) providing people with verbal information about a robot's perceptual capability might not significantly help them predict its behavior.

The DREAM Dataset: Supporting a data-driven study of autism spectrum disorder and robot enhanced therapy.

We present a dataset of behavioral data recorded from 61 children diagnosed with Autism Spectrum Disorder (ASD). The data was collected during a large-scale evaluation of Robot Enhanced Therapy (RET). The dataset covers over 3000 therapy sessions and more than 300 hours of therapy. Half of the children interacted with the social robot NAO supervised by a therapist. The other half, constituting a control group, interacted directly with a therapist. Both groups followed the Applied Behavior Analysis (ABA) protocol. Each session was recorded with three RGB cameras and two RGBD (Kinect) cameras, providing detailed information of children's behavior during therapy. This public release of the dataset comprises body motion, head position and orientation, and eye gaze variables, all specified as 3D data in a joint frame of reference. In addition, metadata including participant age, gender, and autism diagnosis (ADOS) variables are included. We release this data with the hope of supporting further data-driven studies towards improved therapy methods as well as a better understanding of ASD in general.

Folk-Psychological Interpretation of Human vs. Humanoid Robot Behavior: Exploring the Intentional Stance toward Robots.

People rely on shared folk-psychological theories when judging behavior. These theories guide people's social interactions and therefore need to be taken into consideration in the design of robots and other autonomous systems expected to interact socially with people. It is, however, not yet clear to what degree the mechanisms that underlie people's judgments of robot behavior overlap or differ from the case of human or animal behavior. To explore this issue, participants (N = 90) were exposed to images and verbal descriptions of eight different behaviors exhibited either by a person or a humanoid robot. Participants were asked to rate the intentionality, controllability and desirability of the behaviors, and to judge the plausibility of seven different types of explanations derived from a recently proposed psychological model of lay causal explanation of human behavior. Results indicate: substantially similar judgments of human and robot behavior, both in terms of (1a) ascriptions of intentionality/controllability/desirability and in terms of (1b) plausibility judgments of behavior explanations; (2a) high level of agreement in judgments of robot behavior - (2b) slightly lower but still largely similar to agreement over human behaviors; (3) systematic differences in judgments concerning the plausibility of goals and dispositions as explanations of human vs. humanoid behavior. Taken together, these results suggest that people's intentional stance toward the robot was in this case very similar to their stance toward the human.

Exploring the Multi-Layered Affordances of Composing and Performing Interactive Music with Responsive Technologies.

The question motivating the work presented here, starting from a view of music as embodied and situated activity, is how can we account for the complexity of interactive music performance situations. These are situations in which human performers interact with responsive technologies, such as sensor-driven technology or sound synthesis affected by analysis of the performed sound signal. This requires investigating in detail the underlying mechanisms, but also providing a more holistic approach that does not lose track of the complex whole constituted by the interactions and relationships of composers, performers, audience, technologies, etc. The concept of affordances has frequently been invoked in musical research, which has seen a "bodily turn" in recent years, similar to the development of the embodied cognition approach in the cognitive sciences. We therefore begin by broadly delineating its usage in the cognitive sciences in general, and in music research in particular. We argue that what is still missing in the discourse on musical affordances is an encompassing theoretical framework incorporating the sociocultural dimensions that are fundamental to the situatedness and embodiment of interactive music performance and composition. We further argue that the cultural affordances framework, proposed by Rietveld and Kiverstein (2014) and recently articulated further by Ramstead et al. (2016) in this journal, although not previously applied to music, constitutes a promising starting point. It captures and elucidates this complex web of relationships in terms of shared landscapes and individual fields of affordances. We illustrate this with examples foremost from the first author's artistic work as composer and performer of interactive music. This sheds new light on musical composition as a process of construction-and embodied mental simulation-of situations, guiding the performers' and audience's attention in shifting fields of affordances. More generally, we believe that the theoretical perspectives and concrete examples discussed in this paper help to elucidate how situations-and with them affordances-are dynamically constructed through the interactions of various mechanisms as people engage in embodied and situated activity.

The body of knowledge: On the role of the living body in grounding embodied cognition.

Embodied cognition is a hot topic in both cognitive science and AI, despite the fact that there still is relatively little consensus regarding what exactly constitutes 'embodiment'. While most embodied AI and cognitive robotics research views the body as the physical/sensorimotor interface that allows to ground computational cognitive processes in sensorimotor interactions with the environment, more biologically-based notions of embodied cognition emphasize the fundamental role that the living body - and more specifically its homeostatic/allostatic self-regulation - plays in grounding both sensorimotor interactions and embodied cognitive processes. Adopting the latter position - a multi-tiered affectively embodied view of cognition in living systems - it is further argued that modeling organisms as layered networks of bodily self-regulation mechanisms can make significant contributions to our scientific understanding of embodied cognition.

Affordance processing in segregated parieto-frontal dorsal stream sub-pathways.

The concept of affordances indicates "action possibilities" as characterized by object properties the environment provides to interacting organisms. Affordances relate to both perception and action and refer to sensory-motor processes emerging from goal-directed object interaction. In contrast to stable properties, affordances may vary with environmental context. A sub-classification into stable and variable affordances was proposed in the framework of the ROSSI project (Borghi et al., 2010; Borghi and Riggio, 2015, 2009). Here, we present a coordinate-based meta-analysis of functional imaging studies on object interaction targeting consistent anatomical correlates of these different types of affordances. Our review revealed the existence of two parallel (but to some extent overlapping) functional pathways. The network for stable affordances consists of predominantly left inferior parietal and frontal cortices in the ventro-dorsal stream, whereas the network for variable affordances is localized preferentially in the dorso-dorsal stream. This is in line with the proposal of differentiated affordances: stable affordances are characterized by the knowledge of invariant object features, whereas variable affordances underlie adaptation to changing object properties.

Embodied cognition and circular causality: on the role of constitutive autonomy in the reciprocal coupling of perception and action.

The reciprocal coupling of perception and action in cognitive agents has been firmly established: perceptions guide action but so too do actions influence what is perceived. While much has been said on the implications of this for the agent's external behavior, less attention has been paid to what it means for the internal bodily mechanisms which underpin cognitive behavior. In this article, we wish to redress this by reasserting that the relationship between cognition, perception, and action involves a constitutive element as well as a behavioral element, emphasizing that the reciprocal link between perception and action in cognition merits a renewed focus on the system dynamics inherent in constitutive biological autonomy. Our argument centers on the idea that cognition, perception, and action are all dependent on processes focussed primarily on the maintenance of the agent's autonomy. These processes have an inherently circular nature-self-organizing, self-producing, and self-maintaining-and our goal is to explore these processes and suggest how they can explain the reciprocity of perception and action. Specifically, we argue that the reciprocal coupling is founded primarily on their endogenous roles in the constitutive autonomy of the agent and an associated circular causality of global and local processes of self-regulation, rather than being a mutual sensory-motor contingency that derives from exogenous behavior. Furthermore, the coupling occurs first and foremost via the internal milieu realized by the agent's organismic embodiment. Finally, we consider how homeostasis and the related concept of allostasis contribute to this circular self-regulation.

A novel approach to locomotion learning: Actor-Critic architecture using central pattern generators and dynamic motor primitives.

In this article, we propose an architecture of a bio-inspired controller that addresses the problem of learning different locomotion gaits for different robot morphologies. The modeling objective is split into two: baseline motion modeling and dynamics adaptation. Baseline motion modeling aims to achieve fundamental functions of a certain type of locomotion and dynamics adaptation provides a "reshaping" function for adapting the baseline motion to desired motion. Based on this assumption, a three-layer architecture is developed using central pattern generators (CPGs, a bio-inspired locomotor center for the baseline motion) and dynamic motor primitives (DMPs, a model with universal "reshaping" functions). In this article, we use this architecture with the actor-critic algorithms for finding a good "reshaping" function. In order to demonstrate the learning power of the actor-critic based architecture, we tested it on two experiments: (1) learning to crawl on a humanoid and, (2) learning to gallop on a puppy robot. Two types of actor-critic algorithms (policy search and policy gradient) are compared in order to evaluate the advantages and disadvantages of different actor-critic based learning algorithms for different morphologies. Finally, based on the analysis of the experimental results, a generic view/architecture for locomotion learning is discussed in the conclusion.

On the importance of a rich embodiment in the grounding of concepts: perspectives from embodied cognitive science and computational linguistics.

The recent trend in cognitive robotics experiments on language learning, symbol grounding, and related issues necessarily entails a reduction of sensorimotor aspects from those provided by a human body to those that can be realized in machines, limiting robotic models of symbol grounding in this respect. Here, we argue that there is a need for modeling work in this domain to explicitly take into account the richer human embodiment even for concrete concepts that prima facie relate merely to simple actions, and illustrate this using distributional methods from computational linguistics which allow us to investigate grounding of concepts based on their actual usage. We also argue that these techniques have applications in theories and models of grounding, particularly in machine implementations thereof. Similarly, considering the grounding of concepts in human terms may be of benefit to future work in computational linguistics, in particular in going beyond "grounding" concepts in the textual modality alone. Overall, we highlight the overall potential for a mutually beneficial relationship between the two fields.

Toward metabolic robotics: insights from modeling embodied cognition in a biomechatronic symbiont.

We present a novel example of a biomechatronic hybrid system. The living component of the system, embedded within microbial fuel cells, relies on the availability of food and water in order to produce electrical energy. The latter is essential to the operations of the mechatronic component, responsible for finding and collecting food and water, and for the execution of work. In simulation, we explore the behavioral and cognitive consequences of this symbiotic relation. In particular we highlight the importance of the integration of sensorimotor and metabolic signals within an evolutionary perspective, in order to create sound cognitive living technology.

Humanoids Learning to Walk: A Natural CPG-Actor-Critic Architecture.

The identification of learning mechanisms for locomotion has been the subject of much research for some time but many challenges remain. Dynamic systems theory (DST) offers a novel approach to humanoid learning through environmental interaction. Reinforcement learning (RL) has offered a promising method to adaptively link the dynamic system to the environment it interacts with via a reward-based value system. In this paper, we propose a model that integrates the above perspectives and applies it to the case of a humanoid (NAO) robot learning to walk the ability of which emerges from its value-based interaction with the environment. In the model, a simplified central pattern generator (CPG) architecture inspired by neuroscientific research and DST is integrated with an actor-critic approach to RL (cpg-actor-critic). In the cpg-actor-critic architecture, least-square-temporal-difference based learning converges to the optimal solution quickly by using natural gradient learning and balancing exploration and exploitation. Futhermore, rather than using a traditional (designer-specified) reward it uses a dynamic value function as a stability indicator that adapts to the environment. The results obtained are analyzed using a novel DST-based embodied cognition approach. Learning to walk, from this perspective, is a process of integrating levels of sensorimotor activity and value.

Theories and computational models of affordance and mirror systems: an integrative review.

Neuroscientific and psychological data suggest a close link between affordance and mirror systems in the brain. However, we still lack a full understanding of both the individual systems and their interactions. Here, we propose that the architecture and functioning of the two systems is best understood in terms of two challenges faced by complex organisms, namely: (a) the need to select among multiple affordances and possible actions dependent on context and high-level goals and (b) the exploitation of the advantages deriving from a hierarchical organisation of behaviour based on actions and action-goals. We first review and analyse the psychological and neuroscientific literature on the mechanisms and processes organisms use to deal with these challenges. We then analyse existing computational models thereof. Finally we present the design of a computational framework that integrates the reviewed knowledge. The framework can be used both as a theoretical guidance to interpret empirical data and design new experiments, and to design computational models addressing specific problems debated in the literature.

The feeling of action tendencies: on the emotional regulation of goal-directed behavior.

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.

Modeling the Development of Goal-Specificity in Mirror Neurons.

Neurophysiological studies have shown that parietal mirror neurons encode not only actions but also the goal of these actions. Although some mirror neurons will fire whenever a certain action is perceived (goal-independently), most will only fire if the motion is perceived as part of an action with a specific goal. This result is important for the action-understanding hypothesis as it provides a potential neurological basis for such a cognitive ability. It is also relevant for the design of artificial cognitive systems, in particular robotic systems that rely on computational models of the mirror system in their interaction with other agents. Yet, to date, no computational model has explicitly addressed the mechanisms that give rise to both goal-specific and goal-independent parietal mirror neurons. In the present paper, we present a computational model based on a self-organizing map, which receives artificial inputs representing information about both the observed or executed actions and the context in which they were executed. We show that the map develops a biologically plausible organization in which goal-specific mirror neurons emerge. We further show that the fundamental cause for both the appearance and the number of goal-specific neurons can be found in geometric relationships between the different inputs to the map. The results are important to the action-understanding hypothesis as they provide a mechanism for the emergence of goal-specific parietal mirror neurons and lead to a number of predictions: (1) Learning of new goals may mostly reassign existing goal-specific neurons rather than recruit new ones; (2) input differences between executed and observed actions can explain observed corresponding differences in the number of goal-specific neurons; and (3) the percentage of goal-specific neurons may differ between motion primitives.