Keynote Speakers

Prof. Jeffrey Lockman, PhD.
Department of Psychology, Tulane University, USA.

Individuals possess functional maps of the body. They are able to localize targets on their bodies with their hands regardless of where the target is positioned on the body and regardless of where the hands are positioned in space. This type of functional body map has great adaptive value. It helps individuals to engage in activities centered on the body including feeding, grooming, removing stimuli, tool use and many other basic skills of daily living. Little, however, is known about when or how functional maps of the body map develop in humans. Additionally, such developmental information might inform the design of artificial agents where knowledge of body layout is crucial for effective functioning in an environment.
In this presentation, we focus on the origins of body maps in infants. For this purpose, we have developed new procedures to assess how infants manually locate targets on their bodies. We place a small vibrating buzzer on different parts of the body and examine whether infants can localize that buzzer by reaching to it. In our first study, we tested infants cross-sectionally from 7 to 24 months of age (N=60) and placed buzzers at different locations on the face (forehead, ears, mouth) or arms and (shoulder, elbow, and arms). In our second study, we studied infants longitudinally from 2-7 months of age (N=15) and placed the buzzers at different locations on the face (forehead, mouth, chin, ears) to examine how face maps develop. Results across studies suggest that body maps emerge gradually in infants. Infants succeed at locating targets in the mouth region earlier than they do in other facial regions. Additionally, infants succeed at localizing targets on the hand before they succeed at localizing targets at other regions on the arm. We also found that infants show strong patterns of lateralization when they localize targets: They reach ipsilaterally for targets on the face, but contralaterally for targets on the arm (where ipsilateral reaches are not biomechanically possible). We conclude by considering mechanisms for the emergence of body maps in human infants based on experience and self-touch.

Kevin O’Regan, PhD.
Research scientist, Laboratoire Psychologie de la Perception, Paris Descartes University, Paris, France.

We studied 2-6 month old infants’ responses to vibrotactile stimuli presented to five locations: the forehead, right hand, left hand, right foot, left foot (five conditions). Vibrotactile stimulation was provided by small vibrators that were attached to the infant’s body, one at a time in counterbalanced order. Each trial lasted 35s, after which the experimenter removed the vibrator and attached it to the next location. In a (sixth) baseline condition no vibrator was attached to the infant’s body and spontaneous movements were recorded. Four age groups were compared cross-sectionally (3-, 4-, 5- and 6 month olds) and a group of infants were followed longitudinally from 2- to 6 months of age.

In order to compare limb activity across conditions, we analyzed video recordings of infants’ responses using a movement analysis software which calculated the distances travelled by each limb in the two-dimensional plane of the video display. We also performed qualitative analyses of the infants’ movements, coding the direction of gaze, the cases where the infant touched its own body, and movements towards the vibrator.

Our hypothesis was that before they actually reach for the vibrating target, which, according to previous studies, occurs around 6 months of age, infants would demonstrate emerging knowledge about their body’s configuration by producing specific movement patterns associated with the stimulated body area. Furthermore, based on earlier studies that used conjugate reinforcement, we hypothesized that at 3 months infants would produce general whole-body movement patterns upon stimulation, and that more localized movements would gradually emerge with age.

Results showed that at 3 months, infants responded with an increase in general activity when the vibrator was placed on the body, independently of the vibrator’s location. Topographical awareness of the body seemed to appear around 5 months, with specific responses resulting from stimulation of the upper body and hands emerging first, followed by the differentiation of movement patterns associated with the stimulation of the feet. Qualitative analyses revealed specific movement types reliably associated with each stimulated location by 6 months of age, possibly preparing infants’ ability to actually reach for the vibrating target.

We discuss this result in relation to newborns’ ability to learn specific movement patterns through intersensory contingency, as well as in relation to studies that proposed a different sequential order for the emergence of awareness of different body locations.

Matej Hoffmann, PhD.
Researcher, iCub Facility, Istituto Italiano di Tecnologia, Italy.

Humans and animals seamlessly command their complex bodies in space and concurrently integrate multimodal sensory information. To support these capabilities, it seems that some representation of the body is necessary. In this regard, a number of concepts like body schema and body image were proposed. However, while the field is rich in experimental observations, it is largely lacking mechanisms that explain them. Computational models are scarce and address at the most isolated subsystems. Humanoid robots possess morphologies – physical characteristics as well as sensory and motor apparatus – that are in some respects akin to human bodies and can thus be used to expand the domain of computational modeling by anchoring it to the physical environment and a physical body and allowing for instantiation of complete sensorimotor loops. We present our modeling endeavor in the iCub – a baby humanoid robot with 53 degrees of freedom, two cameras with an anthropomorphic design, joint encoders in every joint, and a whole-body artificial skin. The developmental trajectory capitalizes on learning about the body through self-stimulation or self-touch. This behavior was instantiated in the robot and the data thus collected is fed into biologically motivated learning algorithms (such as self-organizing maps) in order to first obtain analogs of primary tactile and proprioceptive maps (areas 3b and 3a). Later, we explore the contingencies induced in tactile and proprioceptive afference by the self-touch configurations and study how they may give rise to first body models with spatial properties.

Prof. Minoru Asada, PhD.
Research director, Division of Cognitive Neuroscience Robotics, Institute for Academic Initiatives, Osaka University, Japan.

We have implemented a child-like body with 22-DOF upper body and 10-DOF legs, both are compliant owing to a pneumatic drive system.
The whole body could be a platform for developmental studies. In
the talk, we introduce preliminary experiments and discuss several issues for future applications.

Andrew Bremner, PhD.
Reader in psychology and head of department, Goldsmiths University, London, UK.
We still know relatively little about how human infants and children come to perceive their own bodies and the relationship between external events and the body. In the first part of this talk I will report on recent findings from my lab pertaining to how infants and young children come to be able to process the multisensory relationships which specify their own bodies and an embodied environment. I will focus particularly on the origins of representations of the position of one's own hand and the location of touches on the body and in external space. I will then go on to describe another programme of research investigating the development of an ability to tailor movements of the body to achieve specific goals in external space. These latter studies demonstrate developmental trajectories in human infancy whereby purposeful actions become more specialised.

Gianluca Baldassare, PhD.
Researcher, Institute of Cognitive Sciences and Technologies, Italian National Research Council, Rome, Italy. 

Newborn children develop progressively more complex sensorimotor skills by exploring own body and external stimuli surrounding them. It has been proposed that this development is guided by the detection of contingencies between own movements and the consequent multimodal sensory effects. These processes are investigated, both with empirical experiments involving babies and with computational models within the recently funded European project `GOAL-Robots -- Goal-based Open-ended Autonomous Learning Robots'. This work focuses on a preliminary computational model of those processes. The progressive enhancement of the model is expected to support the interpretation of the empirical experiments, and to suggest new ones based on its predictions. Moreover, it is expected to support the design of new open-ended learning robotic controllers. The model is based on three components implementing three learning processes: (a) a component, formed by stacked Kohonen neural networks, supporting the acquisition of increasingly abstract goals based on experienced changes in the environment; (b) a component, formed by an echo-state neural network, supporting the acquisition of the skills able to accomplish the goals; (c) a component, based on a predictor of the accomplishment of the pursued goal, used to measure the improvement of each skill and hence to intrinsically motivate the selection of its goal. The computational model is used as the controller of a simulated agent in a 2D environment, composed of two kinematic 3DoF arms. Multimodal sensory information from proprioception, touch, and vision is used by the system to form goals and guide skill learning. Results of the initial tests of the model are presented, together with their possible implications for the empirical experiments.