Robotic Homunculus: Learning of Artificial Skin Representation in a Humanoid Robot Motivated by Primary Somatosensory Cortex

Hoffmann, Matej; Straka, Zdenek; Farkaš, Igor; Vavrecka, Michal; Metta, Giorgio

doi:10.1109/tcds.2017.2649225

Cited by 35 publications

(30 citation statements)

References 49 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…1) Tactile homunculus (superficial schema): One component or "representation" that seems necessary is the "tactile homunculus" or superficial schema. In Hoffmann et al [31], we have obtained this homuncular representation for one half of the upper body of the iCub humanoid: Local stimulations of the skin surface were fed into a self-organizing map algorithm (SOM) that was additionally constrained such that the sequence of body parts on the output sheet mimicked that from the cortex (area 3b) -see Fig. 7B.…”

Section: Remapping Decomposed Into Modulesmentioning

confidence: 99%

See 1 more Smart Citation

Development of reaching to the body in early infancy: From experiments to robotic models

Hoffmann

Chinn

Somogyi

et al. 2017

2017 Joint IEEE International Conference on Development and Learning and Epigenetic Robotics (ICDL-EpiRob)

Self Cite

View full text Add to dashboard Cite

We have been observing how infants between 3 and 21 months react when a vibrotactile stimulation (a buzzer) is applied to different parts of their bodies. Responses included in particular movement of the stimulated body part and successful reaching for and removal of the buzzer. Overall, there is a pronounced developmental progression from general to specific movement patterns, especially in the first year. In this article we review the series of studies we conducted and then focus on possible mechanisms that might explain what we observed. One possible mechanism might rely on the brain extracting "sensorimotor contingencies" linking motor actions and resulting sensory consequences. This account posits that infants are driven by intrinsic motivation that guides exploratory motor activity, at first generating random motor babbling with self-touch occurring spontaneously. Later goal-oriented motor behavior occurs, with self-touch as a possible effective tool to induce informative contingencies. We connect this sensorimotor view with a second possible account that appeals to the neuroscientific concepts of cortical maps and coordinate transformations. In this second account, the improvement of reaching precision is mediated by refinement of neuronal maps in primary sensory and motor cortices-the homunculi-as well as in frontal and parietal cortical regions dedicated to sensorimotor processing. We complement this theoretical account with modeling on a humanoid robot with artificial skin where we implemented reaching for tactile stimuli as well as learning the "somatosensory homunculi". We suggest that this account can be extended to reflect the driving role of sensorimotor contingencies in human development. In our conclusion we consider possible extensions of our current experiments which take account of predictions derived from both these kinds of models.

show abstract

Section: Remapping Decomposed Into Modulesmentioning

confidence: 99%

“…Arrows illustrate the relationship in orientation between the skin parts and the learned map. From [31].…”

Section: Remapping Decomposed Into Modulesmentioning

confidence: 99%

Development of reaching to the body in early infancy: From experiments to robotic models

Hoffmann

Chinn

Somogyi

et al. 2017

2017 Joint IEEE International Conference on Development and Learning and Epigenetic Robotics (ICDL-EpiRob)

Self Cite

View full text Add to dashboard Cite

show abstract

“…Lanillos et al used them for constructing a probabilistic body map for self-perception [21]. Besides, [22] employed self-organizing maps for learning inverse-forward kinematics for self-perception whereas [23] and [24] used them for the learning of tactile maps and body image. In comparison, gain-field networks can combine advantageously the topological self-organization property of SOM with autoencoding and the nonlinear probabilistic mapping property of Bayesian networks based on multiplication.…”

Section: Introductionmentioning

confidence: 99%

Visual Learning for Reaching and Body-Schema with Gain-Field Networks

Abrossimoff

Pitti

Gaussier

2018

2018 Joint IEEE 8th International Conference on Development and Learning and Epigenetic Robotics (ICDL-EpiRob)

View full text Add to dashboard Cite

Perceiving our own body posture improves the way we move dynamically and reversely, motion coordination serves to learn better the position of our own body. Following this idea, we present a neural architecture toward reaching movements and body self-perception from a developmental perspective. Our framework is based on the neurobiological mechanism known as gain modulation in parietal neurons that is found to integrate the visual, motor and proprioceptive information through product-like processes. These multiplicative networks have interesting properties for learning nonlinear transformations such as the head-centered mapping in reaching tasks or the hand-centered mapping for a body-centered representation. In a simulation of a three-link arm, we perform experiments of nearby and far reach targets exploiting one or the other strategy. The later combination of the two networks generates autonomous control toward the target by processing the bodycentered spatial information and the preferred visual direction for the desired motor commands.

show abstract

“…The DAC architecture generates aspects of the ecological self through interoceptive processes that maintain a model of the robotÕs physical parts and the geometry of its current body pose, and exteroceptive processes that monitor the robotÕs immediate surroundings. For example, using somatotopic maps modelled on human primary sensory cortex, and techniques such as self touch, Giorgio Metta, Matej Hoffmann and colleagues have developed methods that allow the iCub to learn its own body model [26], and recalibrate its knowledge of its own geometry [54]. Additionally, by combining vision with tactile sensing and with proprioception, iCub is able to develop a sense of peripersonal space that allows it to predict contacts with objects before they happen [53].…”

Section: A Biomimetic Cognitive Architecture For the Robot Selfmentioning

confidence: 99%

The Synthetic Psychology of the Self

Prescott

Camilleri

2018

Intelligent Systems, Control and Automation: Science and Engineering

View full text Add to dashboard Cite

Synthetic psychology describes the approach of "understanding through building" applied to the human condition. In this chapter, we consider the specific challenge of synthesizing a robot "sense of self". Our starting hypothesis is that the human self is brought into being by the activity of a set of transient self-processes instantiated by the brain and body. We propose that we can synthesize a robot self by developing equivalent subsystems within an integrated biomimetic cognitive architecture for a humanoid robot. We begin the chapter by motivating this work in the context of the criteria for recognizing other minds, and the challenge of benchmarking artificial intelligence against human, and conclude by describing efforts to create a sense of self for the iCub humanoid robot that has ecological, temporally-extended, interpersonal and narrative components set within a multi-layered model of mind. Alan Turing, one of the founders of computer science, once suggested that there were two paths to human-level Artificial Intelligence (AI)Ñone through emulating the more abstract abilities of the human mind, such as chess playing, the other, much closer to the spirit of this book, by providing a robot with Òthe best sense organs that money can buy, and then teach[ing] it to understand and speak English. This process could follow the normal teaching of a childÓ [66, p.460]. Turing was noncommittal about which approach would work best and suggested we try both. Two-thirds of a century after Turing, as different AIs battle between themselves to be the worldÕs best at chess [59], it is clear that the first approach has been spectacularly successful at producing some forms of machine intelligence, though not at emulating or approaching Ògeneral intelligenceÓÑthe

show abstract

Robotic Homunculus: Learning of Artificial Skin Representation in a Humanoid Robot Motivated by Primary Somatosensory Cortex

Cited by 35 publications

References 49 publications

Development of reaching to the body in early infancy: From experiments to robotic models

Development of reaching to the body in early infancy: From experiments to robotic models

Visual Learning for Reaching and Body-Schema with Gain-Field Networks

The Synthetic Psychology of the Self

Contact Info

Product

Resources

About