Search

Inferring others' preferences is one of the central cognitive tasks of social life. The inferred preferences are used to predict the behavior of others in novel situations; therefore, a certain level of abstraction is required. We considered the inference of preferences as a classification problem on an abstract latent feature space. Participants (n=96) were asked to show 89 images of automobiles to either an agent that had been programmed with a specific person's preference and which showed facial expressions according to that preference or an agent that did not show facial expressions. Next, participants were asked to predict the agent's likes and dislikes for novel 23 automobiles. The results showed that participants could predict the agent's likes and dislikes for the unseen automobiles with 74% accuracy, indicating that they learned the decision boundary of agents' preferences on the latent feature space using expressions of emotion as labels.

This paper describes our model of the emergence of a mirror system for imitative learning. The mirror system is an intermodal mapping between someone's action (as seen) and one's own action (to execute). It is widely assumed that the mapping is a geometric transformation between bodies in the visual and motor spaces; however, it is still unclear if the mapping is innate or, if not, how it is acquired. We claim here that the mapping is not a geometric transformation between bodies but a functional correspondence between someone's action on an object for producing an effect and one's own action on the same or similar object for producing a similar effect, which is learnable because both tend to utilize the object's affordance in a similar way.

The paper proposes a computationally feasible method for measuring context-sensitive semantic distance between words. The distance is computed by adaptive scaling of a semantic space. In the semantic space, each word in the vocabulary V is represented by a multi-dimensional vector which is obtained from an English dictionary through a principal component analysis. Given a word set C which specifies a context for measuring word distance, each dimension of the semantic space is scaled up or down according to the distribution of C in the semantic space. In the space thus transformed, distance between words in V becomes dependent on the context C. An evaluation through a word prediction task shows that the proposed measurement successfully extracts the context of a text., Comment: 8 pages, single LaTeX file

In this study we investigated the effects of different morphological configurations on a robot swinging task using van der Pol oscillators. The task was examined using two separate degrees of freedom (DoF), both in the presence and absence of neural entrainment. Neural entrainment stabilises the system, reduces time-to-steady state and relaxes the requirement for a strong coupling with the environment in order to achieve mechanical entrainment. It was found that staged release of the distal DoF does not have any benefits over using both DoF from the onset of the experimentation. On the contrary, it is less efficient, both with respect to the time needed to reach a stable oscillatory regime and the maximum amplitude it can achieve. The same neural architecture is successful in achieving neuromechanical entrainment for a robotic walking task.

How does an individual use the knowledge acquired through self exploration as a manipulable model through which to understand others and benefit from their knowledge? How can developmental and social learning be combined for their mutual benefit? In this paper we review a hierarchical architecture (HAMMER) which allows a principled way for combining knowledge through exploration and knowledge from others, through the creation and use of multiple inverse and forward models. We describe how Bayesian Belief Networks can be used to learn the association between a robot’s motor commands and sensory consequences (forward models), and how the inverse association can be used for imitation. Inverse models created through self exploration, as well as those from observing others can coexist and compete in a principled unified framework, that utilises the simulation theory of mind approach to mentally rehearse and understand the actions of others.

We present a Perception-Action architecture and experiments to simulate imprinting—the establishment of strong attachment links with a “caregiver”—in a robot. Following recent theories, we do not consider imprinting as rigidly timed and irreversible, but as a more flexible phenomenon that allows for further adaptation as a result of reward-based learning through experience. Our architecture reconciles these two types of perceptual learning traditionally considered as different and even incompatible. After the initial imprinting, adaptation is achieved in the context of a history of “affective” interactions between the robot and a human, driven by “distress” and “comfort” responses in the robot.

A method is presented for adapting the sensors of a robot to its current environment and to learn motion flow detection by observing the informational relations between sensors and actuators. Examples are shown where the robot learns to detect motion flow from sensor data generated by its own movement.

This paper focuses on exploring how learning and development can be structured in synthetic (robot) systems. We present a developmental assembler for constructing reusable and temporally extended actions in a sequence. The discussion adopts the traditions of dynamic pattern theory in which behavior is an artifact of coupled dynamical systems with a number of controllable degrees of freedom. In our model, the events that delineate control decisions are derived from the pattern of (dis)equilibria on a working subset of sensorimotor policies. We show how this architecture can be used to accomplish sequential knowledge gathering and representation tasks and provide examples of the kind of developmental milestones that this approach has already produced in our lab.

This paper introduces a control architecture for the learning of complex sequence of gestures applied to autonomous robots. The architecture is designed to exploit the robot internal sensory-motor dynamics generated by visual, proprioceptive, and predictive informations in order to provide intuitive behaviors in the purpose of natural interactions with humans.

We propose ongoing emergence as a core concept in epigenetic robotics. Ongoing emergence refers to the continuous development and integration of new skills and is exhibited when six criteria are satisfied: (1) continuous skill acquisition, (2) incorporation of new skills with existing skills, (3) autonomous development of values and goals, (4) bootstrapping of initial skills, (5) stability of skills, and (6) reproducibility. In this paper we: (a) provide a conceptual synthesis of ongoing emergence based on previous theorizing, (b) review current research in epigenetic robotics in light of ongoing emergence, (c) provide prototypical examples of ongoing emergence from infant development, and (d) outline computational issues relevant to creating robots exhibiting ongoing emergence.

We used a multilevel growth model to describe the developmental trajectories of infant’s coordinated attention between people and objects between 7 and 10 months of age. Additionally, we assed whether the coordinated attention looks were accompanied by smiles as infants interacted social partners. These results confirm the emergence of visual joint attention skills before the end of the first year. These results will be useful in the construction of robotic systems that engage in joint attention.

This article reviews two reflexive motor patterns in humans: Primitive reflexes and motor primitives. Both terms coexist in the literature of motor development and motor control, yet they are not synonyms. While primitive reflexes are a part of the temporary motor repertoire in early ontogeny, motor primitives refer to sets of motor patterns that are considered basic units of voluntary motor control thought to be present throughout the life-span. The article provides an overview of the anatomy and neurophysiology of human reflexive motor patterns to elucidate that both concepts are rooted in architecture of the spinal cord. I will advocate that an understanding of the human motor system that encompasses both primitive reflexes and motor primitives as well as the interaction with supraspinal motor centers will lead to an appreciation of the richness of the human motor repertoire, which in turn seems imperative for designing epigenetic robots and highly adaptable human machine interfaces.

Children with autism have difficulties in social interaction with other people and much attention in recent years has been directed to robots as therapy tools. We studied the social interaction between children with autism and robots longitudinally to observe developmental changes in their performance. We observed children at a special school for six months and analyzed their performance with robots. The results showed that two children adapted to the experimental situations and developed interaction with the robots. This suggests that they changed their interaction with the robots from an object-like one into an agentlike one.

This paper presents results from experimental studies on early language acquisition in infants and attempts to interpret the experimental results within the framework of the Ecological Theory of Language Acquisition (ETLA) recently proposed by (Lacerda et al., 2004a). From this perspective, the infant’s first steps in the acquisition of the ambient language are seen as a consequence of the infant’s general capacity to represent sensory input and the infant’s interaction with other actors in its immediate ecological environment. On the basis of available experimental evidence, it will be argued that ETLA offers a productive alternative to traditional descriptive views of the language acquisition process by presenting an operative model of how early linguistic function may emerge through interaction.

We describe a set of experiments investigating the role of natural language symbols in scaffolding situated action. Agents are evolved to respond appropriately to commands in order to perform simple tasks. We explore three different conditions, which show a significant advantage to the re-use of a public symbol system, through self-cueing leading to qualitative changes in performance. This is modelled by looping spoken output via environment back to heard input. We argue this work can be linked to, and sheds new light on, the account of self-directed speech advanced by the developmental psychologist Vygotsky in his model of the development of higher cognitive function.

Humans use a set of exploratory procedures to examine object properties through grasping and touch. Our goal is to exploit similar methods with a humanoid robot to enable developmental learning about manipulation. We use a compliant robot hand to find objects without prior knowledge of their presence or location, and then tap those objects with a finger. This behavior lets the robot generate and collect samples of the contact sound produced by impact with that object. We demonstrate the feasibility of recognizing objects by their sound, and relate this to human performance under situations analogous to that of the robot.

Social robots recognize and respond to human social cues with appropriate behaviors. Social robots, and the technology used in their construction, can be unique tools in the study of abnormal social development. Autism is a pervasive developmental disorder that is characterized by social and communicative impairments. Based on three years of integration and immersion with a clinical research group which performs more than 130 diagnostic evaluations of children for autism per year, this paper discusses how social robots will make an impact on the ways in which we diagnose, treat, and understand autism.

This paper elaborates on the workplan of an initiative in embodied cognition: RobotCub. Our goal here is to provide background and to motivate our long-term plan of empirical research including brain and robotic sciences following the principles of epigenetic robotics.

In this paper, we propose a model to develop robots’ covert perceptual capability using reinforcement learning. Covert perceptual behavior is treated as action selected by a motivational system. We apply this model to vision-based navigation. The goal is to enable a robot to learn road boundary type. Instead of dealing with problems in controlled environments with a low-dimensional state space, we test the model on images captured in non-stationary environments. Incremental Hierarchical Discriminant Regression is used to generate states on the fly. Its coarse-to-fine tree structure guarantees real-time retrieval in high-dimensional state space. K Nearest-Neighbor strategy is adopted to further reduce training time complexity.

A model of gaze control is describes that includes mechanisms for predictive control using a forward model and event driven expectations of target behavior. The model roughly undergoes stages similar to those of human infants if the influence of the predictive systems is gradually increased.

We describe sensorimotor phase-plots constructed using information theoretical methods from raw sensor data as a way for a robotic agent to characterise its interactions and interaction history. Measurements of the position and shape of the trajectories, including fractal dimension, can be used to characterise the agent-environment interaction.

We present here an example of incremental learning between two computational models dealing with different modalities: a model allowing to switch spatial visual attention and a model allowing to learn the ordinal sequence of phonetical numbers. Their merging via a common reward signal allows anyway to produce a cardinal counting behaviour that can be implemented on a robot.

This poster outlines an Ecological Theory of Language Acquisition (ETLA). The theory views the early phases of the language acquisition process as an emergent consequence of the interaction between the infant and its linguistic environment. The newborn infant is considered to be linguistically and phonetically naïve but endowed with the ability to register a wide range of multi-sensory inputs along with the ability to detect similarity between the multi-sensory stimuli it is exposed to. The initial steps of the language acquisition process are explained as unintended and inevitable consequences of the infant’s multisensory interaction with the adult. The theoretical model deriving from ETLA is tested using the experimental data presented in the two additional contributions from our research team (Gustavsson et al, “Integration of audiovisual information in 8-months-old infants”; Lacerda, Marklund et al. “On the linguistic implications of context-bound adult-infant interactions”). The generality of the ETLA’s concept is likely to be of significance for a wide range of scientific areas, like robotics, where a central issue concerns addressing general problems of how organisms or systems might develop the ability to tap on the structure of the information embedded in their operating environments.

Synchrony detection between different sensory and/or motor channels appears critically important for young infant learning and cognitive development. For example, empirical studies demonstrate that audio-visual synchrony aids in language acquisition. In this paper we compare these infant studies with a model of synchrony detection based on the Hershey and Movellan (2000) algorithm augmented with methods for quantitative synchrony estimation. Four infant-model comparisons are presented, using audio-visual stimuli of increasing complexity. While infants and the model showed learning or discrimination with each type of stimuli used, the model was most successful with stimuli comprised of one audio and one visual source, and also with two audio sources and a dynamic-face visual motion source. More difficult for the model were stimuli conditions with two motion sources, and more abstract visual dynamics—an oscilloscope instead of a face. Future research should model the developmental pathway of synchrony detection. Normal audio-visual synchrony detection in infants may be experience-dependent (e.g., Bergeson, et al., 2004).

What could a robot learn in one day? This paper describes the DayOne project, an endeavor to build an epigenetic robot that can bootstrap from a very rudimentary state to relatively sophisticated perception of objects and activities in a matter of hours. The project is inspired by the astonishingly rapidity with which many animals such as foals and lambs adapt to their surroundings on the first day of their life. While such plasticity may not be a sufficient basis for long-term cognitive development, it may be at least necessary, and share underlying infrastructure. This paper suggests that a sufficiently flexible perceptual system begins to look and act like it contains cognitive structures.

This paper presents an ontogenetic model of self-organization for robotic intermediary vision. Two mechanisms are under concern. First, the development of low-level local feature detectors that perform a piecewise categorization of the sensory signal. Second, the hierarchical grouping of these local features in a holistic perception. While the grouping mechanism is expressed as a classical agglomerative clustering, underlying similarity measures are not pre-given but developed from the signal statistics.

Grounding symbolic representations in perception is a key and difficult issue for artificial intelligence. The ”Talking Heads” experiment (Steels and Kaplan, 2002) explores an interesting coupling between grounding and social learning of language. In the first version of this experiment, two cameras were interacting in a simplified visual environment made of colored shapes on a white board and they developed a shared, grounded lexicon. We present here the beginning of a new experiment which is an extension of the original one with two autonomous robots instead of two cameras and a complex and unconstrained visual environment. We review the difficulties raised specifically by the embodiment of the agents and propose some directions to address these questions.

As in the previous editions, this workshop is trying to be a forum for multi-disciplinary research ranging from developmental psychology to neural sciences (in its widest sense) and robotics including computational studies. This is a two-fold aim of, on the one hand, understanding the brain through engineering embodied systems and, on the other hand, building artificial epigenetic systems. Epigenetic contains in its meaning the idea that we are interested in studying development through interaction with the environment. This idea entails the embodiment of the system, the situatedness in the environment, and of course a prolonged period of postnatal development when this interaction can actually take place. This is still a relatively new endeavor although the seeds of the developmental robotics community were already in the air since the nineties (Berthouze and Kuniyoshi, 1998; Metta et al., 1999; Brooks et al., 1999; Breazeal, 2000; Kozima and Zlatev, 2000). A few had the intuition – see Lungarella et al. (2003) for a comprehensive review – that, intelligence could not be possibly engineered simply by copying systems that are “ready made” but rather that the development of the system fills a major role. This integration of disciplines raises the important issue of learning on the multiple scales of developmental time, that is, how to build systems that eventually can learn in any environment rather than program them for a specific environment. On the other hand, the hope is that robotics might become a new tool for brain science similarly to what simulation and modeling have become for the study of the motor system. Our community is still pretty much evolving and “under construction” and for this reason, we tried to encourage submissions from the psychology community. Additionally, we invited four neuroscientists and no roboticists for the keynote lectures. We received a record number of submissions (more than 50), and given the overall size and duration of the workshop together with our desire to maintain a single-track format, we had to be more selective than ever in the review process (a 20% acceptance rate on full papers). This is, if not an index of quality, at least an index of the interest that gravitates around this still new discipline.

We investigate how a robot can be provided with an architecture that would enable it to developmentally 'grow-up' and accomplish complex tasks by building on basic built-in capabilities. The paper introduces into the basic principles of AIS and presents experimental results from a real robot. To our knowledge, this is the first implementation of an AIS architecture for controlling a real mobile robot

This paper presents the mechanism of Intelligent Adaptive Curiosity. This is a drive which pushes the robot towards situations in which it maximizes its learning progress. It makes the robot focus on situations which are neither too predictable nor too unpredictable. This mechanism is a source of self-development for the robot: the complexity of its activity autonomously increases. Indeed, we show that it first spends time in situations which are easy to learn, then shifts progressively its attention to situations of increasing difficulty, avoiding situations in which nothing can be learnt.

In this short paper we describe a robotic setup to study the self-organization of conceptualisation and language. What distinguishes this project from others is that we envision a robot with specic cognitive capacities, but without resorting to any pre-programmed representations or conceptualisations. The key to this all is self-organization and enculturation. We report preliminary results on learning motor behaviours through imitation, and sketch how the language plays a pivoting role in constructing world representations.

In this paper we propose an approach to robot learning by imitation that uses the multimodal inputs of language, vision and motor. In our approach a student robot learns from a teacher robot how to perform three separate behaviours based on these inputs. We considered two neural architectures for performing this robot learning. First, a one-step hierarchial architecture trained with two different learning approaches either based on Kohonen's self-organising map or based on the Helmholtz machine turns out to be inefficient or not capable of performing differentiated behavior. In response we produced a hierarchial architecture that combines both learning approaches to overcome these problems. In doing so the proposed robot system models specific aspects of learning using concepts of the mirror neuron system (Rizzolatti and Arbib, 1998) with regards to demonstration learning.

A computational model with three interacting components for context sensitive reinforcement learning, context processing and automation can autonomously learn a focus attention and a shift attention task. The performance of the model is similar to that of normal children, and when a single parameter is changed, the performance on the two tasks approaches that of autistic children.

While motor skill acquisition process is regarded as development of coordination, typically regarded as synchronisation among joint movements, we found another phenomenon which we call differentiation as a consequence of synchronisation. The synchronised movement established is decomposed into several sections or modulated to be executed on different timings without breaking the coordination among them, resulting in the gain of efficiency or flexibility. In the acquisition of skills, the coordinative structure thus goes through two stages: synchronisation and differentiation. We verify in this paper our observation through our experiments and dynamical analysis of the kneading of ceramic art and playing the shaker in samba.

We explore two controversial hypotheses through robotic implementation: (1) Processes involved in recognition and response are tightly coupled both in their operation and epigenesis; and (2) processes involved in symbol emergence should respect the integrity of recognition and response while exploiting the periodicity of biological motion. To that end, this paper proposes a method of recognizing and generating motion patterns based on nonlinear principal component neural networks that are constrained to model both periodic and transitional movements. The method is evaluated by an examination of its ability to segment and generalize different kinds of soccer playing activity during a RoboCup match.

The results from a series of perception experiments designed to test 8-month-old infants’ ability to derive linguistic information from audio-visual events are reported in this presentation. Using a visual preference technique, groups of 8-month-old infants were tested on their ability to extract linguistic information implicit in short video sequences where the images displayed different puppets and the audio tracks presented sentences describing the puppets in naturalistic infant-directed speech style. To assess the relative importance of memory and attention factors, the prosodic and syntactic structure of the speech materials was systematically changed across different groups of subjects. The experimental results are interpreted in terms of the emergentistic acquisition model discussed in the paper presented by Lacerda et al. (“Ecological theory of language acquisition”).

We present a framework for addressing a challenging trade-off between influencing the learning of a robot through design and through social interactions. We identify different kinds of influences that a designer can introduce at design time, and that an expert can introduce using social interactions, and we use these to characterise a two-dimensional design space. As well as discussing how the two sources of influence affect each other, we propose how learning performance typically varies as a result, and present some empirical findings.

This paper discusses the concept of joint attention and the different skills underlying its development. We argue that joint attention is much more than gaze following or simultaneous looking because it implies a shared intentional relation to the world. The current state-of-the-art in robotic and computational models of the different prerequisites of joint attention is discussed in relation with a developmental timeline drawn from results in child studies.

This paper presents developmental learning on a humanoid robot from human-robot interactions. We consider in particular teaching humanoids as children during the child's Separation and Individuation developmental phase (Mahler, 1979). Cognitive development during this phase is characterized both by the child's dependence on her mother for learning while becoming awareness of her own individuality, and by self-exploration of her physical surroundings. We propose a learning framework for a humanoid robot inspired on such cognitive development.

For several decades experiments have been performed where animals have been reared in environments with orientationally restricted contours. The aim has been to find out what effects the visual field has on the development of the visual system in the brain. In this paper we describe similar experiments performed with a robot acting in an environment with only vertical contours and compare the results with the same robot in an ordinary office environment. Using metric projections of the informational distances between sensors it is shown that all visual sensors in the same vertical column are clustered together in the environment with only vertical contours. We also show how the informational structure of the sensors unfold when the robot moves from the environment with oriented contours to a normal environment.

Binding is one of the most fundamental cognitive functions, how to find the correspondence of sensations between different modalities such as vision and touch. Without a priori knowledge on this correspondence, binding is regarded to be a formidable issue for a robot since it often perceives multiple physical phenomena in its different modal sensors, therefore it should correctly match the foci of attention in different modalities that may have multiple correspondences each other. We suppose that learning the multimodal representation of the body should be the first step toward binding since the morphological constraints in self-body-observation would make the binding problem tractable. The multimodal sensations are expected to be constrained in perceiving own body so as to configurate the unique parts of the multiple correspondence reflecting its morphology. In this paper, we propose a method to match the foci of attention in vision and touch through the unique association by cross-anchoring different modalities. Simple experiments show the validity of the proposed method.