A Neural Network Model of the Connectivity of the Biological Somatic Sensors

Introduction

1.0 The Starting Point

The starting point for a design of a sensory model is the modalities of the various sensors of the biological sensory systems. In medical textbooks [1] and most neuroscience textbooks [2], [3], [4], the modalities of sensors are defined in terms of the “conscious” sensation that they evoke. The law of specific nerve energy or “labeled line” principle is often used to explain the unique “conscious” sensation that each modality generates [1], [5]. The law of specific nerve energy ensures that each type of sensor responds specifically to the appropriate form of stimulus that gives rise to a specific sensation. Furthermore, the specificity of each modality is maintained in the central connections of sensory axons, so that stimulus modality is represented by receptors, afferent axons and central pathways that it activates [5]. In the following sections this connectivity, the so called modality-specificity, is reverse engineered from the biological receptor that gives rise to the signal to the somatosensory cortex in the central nervous system (CNS).

1.1 Some Functional Characteristics of the Modality-Sensation

The cognitive scientist, Steven Pinker [6], refers to Ray Jackendoff [7] and Ned Block [8], in discussing the four characteristics of “conscious” sensations. These functional characteristics of the mind or brain include

1. Self Knowledge.

2. The mind builds an internal model of the world that includes the “self.”

3. Access to information or “access consciousness” [9].

4. Sentience or “qualia”: The subjective experience of feeling or sensation (includes the feeling of tactile “touch” or tactile “pain”).

The design of the fourth quality of conscious sensations is not the goal of this paper. This paper shall present a reverse engineered model of the biological sensors (that have a modality associated with them) and their connectivity [10]. Our primary concern, once an engineering design is obtained, is to generalize the design, and make predictions that may be observed and tested. Tests are performed to determine whether the functional characteristics of the sensory model are consistent with the utility and functions of the biological sensors and the modalities associated with them. Functional tests are performed on the first three characteristics of “conscious sensation” enumerated above (excluding the fourth characteristic).

Figure 1: Transforming the cortical folds in the brain into a 3-dimensional external-mirror nodal-mapping.

1.2 The Reverse Engineering Problems Associated with the Design of the Sensory Circuit Model

The sensory model consists of the reverse-engineered design of the connectivity and the neuronal pathways of the biological somatic sensors. The sensory model is a neural network circuit model, called a Relational Robotic Controller (RRC)-circuit1. The building path selected for the RRC-model is to design and build a non-biological RRC brain-like controller that has the first three functional characteristics enumerated above. That is, the RRC controller must control a reverse-engineered mechanical robot that emulates the biological somatic motor system and sensors. These functional characteristics translate into four reverse engineering problems:

Problem 1. How to build a neural network that adheres to the second “consciousness” constraint: The mind builds an internal model of the world, a “world-space”-coordinate frame within the RRC, that includes the “self” in the center. The selected approach was to design a “homunculus” in the controller that is defined by the somatic body sensors, and to define the world space around the homunculus by the positions occupied by flailing limbs.

Problem 2. How to give the internal model “self”-knowledge: The selected approach was to assume that “self”-knowledge would be gained by the robot if the robot would be required to learn the location of each of its surface parts with respect to and related to other surface parts. This knowledge was to be gained by learning how to move a robotic limb towards any and every other part of the robotic body. The robot would have the capability to perform an “itch-scratch” response, by moving a limb through a goal-directed, obstacle avoiding trajectory aimed at an “itch”-point on the robotic body.

Problem 3. How to design the robotic controller so that the trajectory of motion is pre-planned and goal-directed with the option of re-planning (obstacle-avoiding) a pre-planned trajectory: In order to achieve obstacle avoidance (see section II-2.2), the selected approach was to divide the trajectory of motion into small transitions to adjacent nodes. During each frame period, the total goal directed trajectory is preplanned by the Controller as a sequence of small transitions. However, only the first small transition is activated by the controller and the maximum speed of operation of the robot is one transition per frame period.

Problem 4. How to train/program the RRC-world-map-circuit to perform “location” and “itch-scratch” type actions: The selected approach was to utilize state of the art techniques for training/programming a neural network.

II Main Results

1.0 A World Map-Coordinate Frame in the Brain

The primary constraint imposed on the design of the world map-coordinate frame is that the topographic ordering of neural network neurons within the controller form a model of the external world and place the homunculus of the “self’ in the center. This constraint may be satisfied by reverse engineering the topographic ordering of neurons in the animal brain.

1.1 The biological world map in the brain is based on the observed topographic ordering of neurons in the brain

In the animal brain, the origin of the sensory signals, the input signals, and the destination of the control signals, the output signals, is in the three dimensional space in which the animal is located. In order to form a neural world-map coordinate frame, a transformation is required of the three dimensional external space into the coordinate frame in the brain. Figure 1 illustrates the transformation of the neuronal folds in the brain into the 3-dimensional external (mirror) nodal map containing the homunculus of the robot. Validation of such transformations may be obtained by reference to most text books in cognitive neural science, see for example Neuroscience [4] (p. 415), [11] (p. 159). These textbooks show experimental observations of the transformation of the neuronal folds of the somatosensory cortex into a three dimensional mirror nodal map space containing the homunculus of a human. The location of neurons in the somatosensory cortex is defined by the somatic sensors that are densely located along the peripheral surface area of the body. The transformation of the neuronal folds of the motor cortex into a three dimensional space containing the homunculus of a human is defined by the distribution of muscles/motors throughout the body. Dale Purves [12], [11] (p.161), refers to a variety of observations over the last 40 years of iterated somatotopic and topographical substructures within the sensory cortices.

1.1.1. Definition of the “self” at the center of the world map-coordinate frame in the brain: The somatic sensory system is different from other sensory systems in that its receptors are distributed throughout the body. Thus the system is well suited for the definition of the boundary between the “self” and the external world. The mechanoreceptors, embedded in the dermis under glabrous and hairy epidermal layers, have different axonal pathways and stimulate different regions of the brain [5] (p 46). The mechanoreceptors that define the boundary of the “self” in the external world may be used to define the “self” and the space around the self, in the brain.

1.2 A world map coordinate frame within the controller

The design of the “self” in the RRC controller takes the form of a somatotopic and topographical substructure made up of receiving neurons of a neural network. The location of neurons in the topographic ordering (in the brain/controller) bear a one to one correspondence with the mechanoreceptors/pressure transducers that are distributed on the skin surface of the body. The mechanoreceptors are reverse-engineered by pressure transducers distributed along the surface of the robotic body. Thus, a pressure transducer signal originating in the external coordinate frame, is connected via the nervous system to a tactile receiving neuron at a corresponding nodal position in the internal nodal map in the brain/controller. The connectivity in the internal nodal map is between the indexed coordinate location and the pressure transducers on the robotic body. The receiving neurons that define the “self” in the brain are shown in Figure 2. For visualization convenience the reverse-engineered input-internal world map in the RRC controller is assumed to be configured in three dimensions similar to the external nodal map distribution of tactile sensors rather than being configured similar to the folds in the brain (see Figure 1). Each node of the three dimensional internal world map is the location of a receiving neuron, receiving data from a tactile body sensor. The origin of the system may be determined by sensory data from the thoracic cavity transformed into a receiving neuron at the center of the internal set of neurons defining the thoracic cavity.

		The motion of all body parts may then be determined relative to the internal center (origin) at the thoracic cavity. The biological near space may be defined by the potential positions of flailing limbs in the external world. The RRC near space may also be defined by flailing mechanical limbs. The near spaces in the internal regions of the brain/controller are shown in Figures 2. A neuron may be assigned to a nodal position in the internal near space even though that position is unoccupied by a flailing limb. The receiving neuron at the corresponding node may have tactile data projected on it when and if the corresponding location in the external near space is occupied by a flailing limb. Regions of the external mirror nodal map unoccupied by flailing limbs are defined in the internal map by dormant receiving neurons (in the brain/controller). The signal originating at each somatic sensor and received by the receiving neuron is designated as the q-input signal. The position-location of flailing limbs, head, and hips may be determined by the q-input signal, relative to the origin of the coordinate system. All motion of robotic parts occurs relative to the fixed location of the thoracic cavity: flailing limbs, head and neck rotational motion, and hip rotational motion.
Figure 2: A neuronal world-map coordinate frame within the brain showing the location of neurons that define the “self” and the near-space around the “self”.

2.0 Measuring “Self” Knowledge: The self-identification and location circuit

Self-knowledge in the RRC-circuit consists of data relating to the location of the various body parts with respect to the origin of the coordinate frame located within the controller (see Figure 2). The robotic controller is then required to learn the location of each of its surface parts with respect to and related to the other surface parts. This knowledge is gained by learning how to move, first a robotic finger and then all other moveable body parts towards any and every other part of the robotic body. In other words, the robot would have the capability to perform an “itch-scratch” response, by moving a limb through a goal directed trajectory aimed at an itch-point, located at a mechanoreceptor (pressure transducer) on the robotic body. The itch-point is defined as the q-final goal position of the trajectory of motion. The robotic control circuit that controls the “itch-scratch” trajectories of motion is called a self-identification and location circuit. These trajectories and the “self knowledge” requirement generate a “known” measure of the self-space and also a “known” measure of the space surrounding the self (the near space).

2.1 The trajectory of the initial position, q-initial, of a robotic finger through the world map-coordinate frame

2.1.1 The biological position measurements by the proprioceptive system: The initial position of a flailing limb is generally determined by signals received from the muscle and joint proprioceptor receptors [5] (p. 258). The perception of limb position and movement is mediated by three main types of peripheral receptor that signal the stationary position of the limb and the speed and direction of limb movement: 1. Mechanoreceptors located in joint capsules: the joint proprioceptors respond to changes in the angle, direction, and velocity of movement of the joint. 2. Muscle spindle receptors: mechanoreceptors in muscles that are specialized for the detection of changes in muscle length (stretch). 3. Cutaneous mechanoreceptors from Golgi tendon organs: monitor muscle tension, or force of contraction. In addition there are 2-sub-modalities of limb proprioception: the sense of stationary position of the limb (limb position sense) and the sense of limb movement (kinesthesia) [2] (p. 346).

2.1.2 The reverse engineered mechanical position measurements: In the reverse-engineered RRC the limb is replaced by a robotic arm and the proprioceptors are replaced by angle measuring transducers. The muscles of the biological arm are reverse-engineered by a robotic arm with three motors at each joint replacing each muscle set. The proprioceptors data are reverse-engineered by angle measuring transducers associated with each angular torque generated by each of the three motors. Figure 3 shows the angle measuring transducers located at the joint-gimbals locations of the shoulder and elbow that simulate the output of the proprioceptors. Figure 4 shows the position of a robotic finger that is measured by the angle measuring transducers. The internal nodal map position of q-initial is a function of the gimbal-angle location reading that is transmitted to the RRC-controller at the rate of one reading per frame period.

	In order to connect a signal originating at a q-initial position of a robotic finger, in the external (mirror) map, to its corresponding point in the near space of the internal nodal map, one must convert the signals received from the angle measuring transducers into nodal position data. In the design of the robotic arm, an intermediate circuit is required that converts the angle measurements to a location in the near space of the internal nodal map, and transmits the q-initial signal to a receiving neuron at that location.
Figure 3: The angle measuring transducers located at the gimbal positions of the shoulder and elbow. The transducers simulate the output of proprioceptors.		Figure 4: The position of a robotic finger that is measured by the angle measuring transducers shown in figure 3.

2.2 The “itch-scratch” trajectory: Control of Motion in the World Map in the Controller

The itch-scratch trajectory must adhere to two fundamental constraints that are derived from the cognitive neuroscience literature. The first is that all motion should be pre-planned and end-point goal directed [2], [3]. The second is a volitional constraint that allows a volitional robot the option of re-planning a pre-planned trajectory whenever an obstacle is detected along the path of the pre-planned trajectory2.

In order to design the RRC so that the trajectory of motion is pre-planned and goal directed with the option of re-planning a pre-planned trajectory, the trajectory of motion is decomposed into small transitions to adjacent nodes. During each frame period the total goal directed trajectory is determined by the RRC as a pre-planned sequence of small transitions. However, during any frame period, only the first small transition is activated by the controller. And the maximum speed of a robotic part along the trajectory is one nodal transition per frame period.

3.0 A summary description of the RRC-system

Given the reverse engineered input and output of the RRC-circuit (see section II-1.1 relating to the biological somatotsensory and motor cortices), it is now possible to specify a building path that is functionally similar to the brain. In this case the q-initial and the goal position, q-final are associated with the somatosensory cortex, and the control p-signals are associated with the motor cortex.

3.1 The Conjunction of the Sensory Input Map with the Motor Control Output Map

The topographic distributions of sensory and motor neurons, shown in Figure 5, form a separate input neural network in the region related to the somatosensory cortex, and a separate output neural network in the region related to the motor cortex. In the reverse-engineered design of the RRC these two topographic distributions are combined into a single Nodal Map Module. The combined sensory input and motor control output is formed by a conjunction of topographic distributions of sensory neurons and motor neurons. Figure 5 shows the combined sensory and motor control internal world map. In the Figure, a control p-signal originates in the motor cortex and is applied to the Nodal Map Module at the corresponding q-initial position. The q-initial signal originates in the external nodal map, and is applied, via the simulated somatosensory cortex, to the corresponding q-initial position in the Nodal Map Module.
The neuronal distribution of p-generating neurons in the internal world map is determined by the distribution of muscles/motors in the external map. However the point of application of each p-signal in the external map, is at the q-initial location of the robotic part controlled by the p-signal.

Figure 5: A flow diagram of the q-vector and p-vector through a RRC that simulates the functionality of the human brain. The output of the Nodal Map Module goes to the external-mirror nodal-map via the Sequence Stepper and the Control Signal-output Modules.

3.2 The Design of the Nodal Map Modules:

The design of the Nodal Map Module proceeds with the requirement that a nodal transition between two adjacent points in the internal nodal map, is a measure of a nodal transition between two adjacent nodes in the three-dimensional Euclidean space external to the robot. A trajectory of motion of a robotic part is controlled by a sequence of nodal transitions in the internal space, occurring at the rate of one nodal transition per frame period (see the sequence of p-q nodal transitions shown in the mirror nodal map in Figure 5).

Each joint in the animal body is associated with a p-q Nodal Map Module. Each Nodal Map Module is associated with a set of motor control neurons in the brain that may control up to three degrees of freedom per muscle joint [13]3. For example, the muscles of the upper arm control the motion of the elbow with respect to the shoulder, whereas the muscle set of the lower arm controls the motion of the wrist with respect to the elbow. In each case, three motors and three angle measuring transducers are required at each joint (as shown in Figure 3). The q-initial location of a robotic part associated with each robotic joint is generated by the angle measuring transducer associated with that joint. A p-q nodal map is required for each joint in the animal body. The q-initial position of the part associated with that joint is applied to the corresponding p-q nodal map.

3.3 The Design of the RRC-system

Figure 6 shows an array of RRC-circuits and the central location of the large number of self location and identification nodal maps that are configured like a world space-coordinate frame. Each Nodal Map Module consists of a conjunction of the q-input and p-output neural networks. The internal nodal maps form a combined control and sensory signal model of the external world and places the homunculus of the robotic-self in the center. Each node of the internal nodal map has a p-signal and q-signal associated with it. The q-signal locates the node in the external nodal map whereas the p-signal causes a displacement to an adjacent node. The flow of the q, p-signals through the block diagram (Figure 6), is illustrated in figure 5. The output of the Nodal Map Module goes to the external-mirror nodal map, via the Sequence Stepper Module and the Control Signal Output Module. For example in Figure 5, the displacement of the wrist, illustrated in the external-mirror nodal map, takes place between a set of three adjacent nodes, from q-initial to q-final.

Figure 6: A hierarchical array of RRCs with the “self”-circuit at the center.

3.4 A Giant Parallel Processor

A complete and separate neural network circuit is required for each moveable part of the robot. The neural networks for the elbow (qi-elbow), wrist (qi-wrist), hand (qi-hand) etc., must operate simultaneously and hierarchically during each frame period. Thus each near space nodal location may contain multiple neurons and different neural networks for each qi-location-neuron (qi-elbow, qi-wrist, qi-hand etc.) [14].

The reverse-engineered brain is a giant parallel processing unit that simultaneously controls all the joints present in the animal body. Any internal representation of motion must accommodate 244-different degrees of freedom that involve more than 600-different body muscles. Excluding all facial muscles and joints there are approximately 65-joints in the human body (34-arm, 26-leg and 5-hip shoulder and neck [14]. Therefore, 65-p-q nodal maps may be required in order to achieve self-knowledge by the locomotive “itch-scratch” control of the human body. It is assumed that the output signals of each internal nodal map consists of sets of signals that control each muscle set used to displace a particular joint-limb3. The total reverse-engineered output map, analogous to the topographic distribution of neurons in the motor cortex, consists of all the neurons associated with all the muscle sets of the body.

3.5 The solution to the equations of motion in the Design of the RRC Circuit

A complete solution to the equations of motion of an RRC- circuit for the sensory-motor control of a robotic arm has been developed by Rosen & Rosen [14]. The solution to the neural net equations represented by thresholds and the synoptic weights of interconnections is shown in Figure 7 [14]. The solution is based on the work of Teuvo Kohonen [15] and Helge Ritter [16]. The RRC-controlled motion of a 29-degree of freedom robotic arm includes the motion of the elbow (qi-elbow) with respect to the shoulder, the motion of the wrist (qi-wrist) with respect to the elbow, the the motion of the knuckles with respect to the wrist (qi-knuckles), and the motion of the fingers with respect to the knuckles. Any q-initial location measurement in the external map triggers only that fiber output that is connected to the corresponding location of the receiving neuron in the internal near space. This circuit continuously monitors the position of only one part of the robot (e.g. elbow, wrist, hand) and traces a path in the mirror map corresponding to the spatial displacements of that one part (see Figure 4).

Figure 7: A pictorial representation of a neural network used to control 3-degrees of freedom. The location of brain neurons in the configured input layer, the “self”, the near-space receiving neurons, and the trajectory between qix and qfx is shown in the Figure. The solution to the neural net equations represented by thresholds and the synoptic weights of interconnections is also shown in the Figure (Rosen & Rosen, 2003d).

4.0 Programming The RRC: Learning “Self” Knowledge

All the nodes of each Nodal Map Module must be trained before it may be used to control the motor actions of the robot. Details for the training of all the nodes of an RRC-circuit, to perform a complete set of “itch-scratch” motor control tasks are described in an article titled “The Engineering Design of a NCC-circuit for the Sensory-motor Control of a Robotic Arm” [14].

Training a nodal map module is accomplished by use of a modified “Hebbian” learning rule [17]. Figure 8 is a training flow diagram of the p-vectors and q-vectors through the RRC during one frame period. Two paths are shown in the Figure, a training path and an operational path. Training is performed on the Nodal Map Module and on the Sequence Stepper Module. The Nodal Map training consists of the tabular assignment of a correct set of p-values to each nodal location of the Nodal Map Module. The correct p-value, a table-line-entry, is that control signal that causes an exact motor displacement of a robotic part, to an adjacent node, in the external map shown in Figure 5. Each table-line entry is made up of 27-correct p-signal transitions to adjacent nodes in a three dimensional nodal map, and 8-p signal transitions to adjacent nodes in a two dimensional nodal map (see Figure 5). The total set of table-line entries assigned to all the nodes of a Nodal Map Module, encompass all the p-signals that lead to exact transitions to all adjacent nodal positions.

The Task Selector Module (TSM) shown in Figures 5, 6 and 8, generates the q-final “itch” location and motivates the robot to perform a “scratch” action aimed at the “itch”-point. For training purposes, the q-final “itch” location may be artificially generated by the TSM. The TSM-activated q-final “itch”-location becomes a Task-initiating Trigger (TT) that activates the Sequence Stepper Module to examine the region in the Nodal Map Module, between q-initial and q-final, and select a pre-planned trajectory between q-initial and q-final.
The method of selecting the correct control signal, p-value, required to move a robotic part from q-initial to q-final, is shown in Figure 8. The correct table-line p-value at the initial node is that value that generates an exact displacement from the initial node at q-initial to the adjacent node at q-final. Correction of the table line entry p-value proceeds by noting the displacement error generated by p-initial and applying a correction factor (shown as CFI in Figure 8) so that p-initial (corrected) leads to an exact transition to the final node defined by q-final (see Figure 8).
The internal nodal map is said to be “trained” if each nodal table, assigned to each node, is made up of a complete set of p-signal line entries (27-p-values in a 3-dimensional nodal map). Each p-signal line entry causes an exact motor displacement to an adjacent external nodal position.
In order to train all the nodes in the nodal space, the q-final-TT is initially selected at nodes that are immediately adjacent to a q-initial. For each sequence stepper and nodal map combination, and for each node defined in a nodal map, the training proceeds with the q-final location placed at distances of two, three, four, and more nodal distances from q-initial. When all the nodes of all the nodal maps are fully trained, the RRC-circuit is said to exhibit “self location and identification” knowledge. The robot has now learned how to move a robotic limb towards any and every part of the robotic body.

Figure 8: Training the nodal map module: Two paths are shown in the Figure, a training path and an operational path. A p-signal is trained and assigned to a node in the Nodal Map Module when CFI is less than delta. Twenty-seven p-signals must be trained at each nodal position.

III. Conclusions

A RRC-circuit that learns all possible itch-scratch trajectories and a self location and identification type of control may be characterized by a form of self knowledge that emulates biological self knowledge.

1.0 Motor Control:

Self-knowledge of the “measure” of the near space (with the self in the center) coupled with self-knowledge (awareness) of the internal reactions of the body to external stimuli, are powerful sensory-motor control tools for all locomotive actions performed by all body parts. The RRC-circuit mediated by the reverse engineered proprioceptors and the vestibular system and coupled with location self-knowledge, operates as a feedback loop that facilitates learning precision control of body motion in an external environment defined by the world map in the brain. The “Location and identification” type of self-knowledge generates a powerful tool for designing and building a volitional humanoid robotic system [18]. In most modern publications relating to sensory-motor control of the somatic motor system, the power of the sensory modalities is rarely taken into consideration and the connectivity associated with it, in the form of a neural representation of a world space in the controller, is never taken into consideration (See for example Guenther et al [19], and Stephen Grossberg [20].

2.0 Perception of the external world: A recording monitor in the brain

The “tactile” sensors of the RRC circuit constantly monitor the external environment of the robot for any tactile contact with the robotic body. The sensors may be viewed as “perceiving” the external environment. The biological sensors, that have a modality associated with them, perceive the external world via a subjective experience of “touch-feeling” or “pain-feeling”. The mechanical sensors, the pressure transducers, may or may not have a “touch-pain” feeling correlated with them. Nonetheless, they “perceive” the external environment by constantly monitoring it for any possible activation, and by responding to each and every activation with “self-location and identification” self-knowledge.
The configured input circuit, the tactile self-circuit defined by the modality-connectivity of the mechanoreceptors distributed on the skin surface of the body, form a recording monitor in the brain. During the life of the robot, whenever a pressure transducer is activated the robot may be said to be “aware” and “conscious” of of the activation. The constant monitoring of the subjective experience of touch-feeling or pain-pressure, that is, “Perception” coupled with “self awareness”, enhances Darwinian survival of the biological organism by supporting the Darwinian search engine that gives the organism knowledge of environmental contingencies, danger, food, shelter and mating opportunities. The RRC circuit and the Darwinian search engine that is used to search the environment for environmental contingencies that impact the survival of the organism, also gives rise to the subjective experience of “seeing”, [21] “hearing,” “smelling,” and “tasting.”4

IV. Notes

1. A RRC-circuit has been designed, reduced to practice and patented on May 6, 2003 (Patent No. US 6,560,512B). All MCon publications may be viewed at www.mcon.org.

2. This is an operational definition of “volition”. For example, if an obstacle appears along the pre-planned “scratch” trajectory of motion, the robot must have the option of re-planning the planned trajectory so as to avoid the obstacle. The robotic controller is said to be a volitional controller if the controlled trajectory of motion is goal directed and pre-planned, with the option available for re-planning the pre-planned trajectory if an environmental contingency is detected prior to reaching the pre-planned goal. Re-planning is always a function of the contingency that appears in the region of the pre-planned path. It is never functionless or random.

3. The simultaneous evolutionary adaptation of the organic muscle-joint and the control system (in the brain) that controls the muscle-joint, implies a one to one relationship between each separate muscle set in the body, and the internal p-q nodal map that is defined for the control of the muscle-joint. Thus, the set of equations that control the muscle set associated with one bodily joint, may also control all the muscle sets associated with all the other body joints.

4. The monitoring function leads to sensory “awareness” or sensory “consciousness”. Tactile monitoring results in “touch” sensations by the “self”-circuit. Visual monitoring results in “seeing” sensation by the “self”- circuit. Auditory monitoring results in “hearing” sensation by the “self”-circuit, etc. In each case the characteristics and connectivity of the mechanism rests on the Darwinian assumption that “self” awareness is an attribute that evolved by Darwinian natural selection in order to fulfill the Darwinian prerogative to survive and reproduce.

V. REFERENCES

[1] A. C. Guyton, Textbook of Medical Physiology. Philadelphia: W.B. Saunders Co., 1991. [2] E. R. Kandel, J. H. Schwarts, & T. M. Jessell, Editors, Principles of Neural Science. Norwalk Conn:Appleton and Lange, 1991. (See also Chapter 26: “Touch”)[3] M. S. Gazzaniga, R. B. Ivry, & G.M. Mangun, Cognitive Neuroscience, New York: W.W. Norton and Co. (Chapter 10 Motor Control) 1998.

[4] M. F. Bear, B.W Connors, & M.A. Paradiso, Neuroscience. 2nd ed. Baltimore Md: Lippincott Williams & Wilkins, 2001.

[5] D. E. Haines, Fundamental Neuroscience 2nd ed. Churchill Livingston: Philadelphia PA, 2002.

[6] S. Pinker, How the Mind Works, New York: WW Norton & Co., 1997.

[7] R. Jackendoff, Consciousness and the Computational Mind, Cambridge Mass: MIT Press, 1987.

[8] N. Block, & commentators. On Confusion About a Function of Consciousness, Behavioral and Brain Sciences, 18, 227-287, 1995

[9] F. Crick, & C. Koch, A framework for Consciousness. Nat. Neurosci. 6, 119-126, (2003).
[10] D. C. Dennett, Artifact Hermeneutics, or reverse Engineering p. 194 in Evolution edited by Mark Ridley Oxford University Press: New York 1997.

[11] D. G. Purves, A. D. Fitzpatrick, L.C. Katz, A. S. La Mantia, & J. O. McNamara, Neuroscience. Sunderland Mass: Sinauer Assoc. Inc., 1997.

[12] D. G. Purves, D. Riddle, & A. LaMantia, Iterated patterns of brain circuitry (or how the cortex gets its spots). Trends Neurosci. 15, 362-368, 1992.

[13] C. DeDuve, Vital Dust: Life as a Cosmic Imperative. New York: basic Books, 1995.

[14] A. Rosen A., & D. B. Rosen, The Engineering Design of an NCC circuit for the Sensory-motor Control of a Robotic Arm. Machine Consciousness Technical Journal. 1, 57-69, 2003 (available for viewing at www.mcon.org)

[15] T. Kohonen, Self Organizing Maps (3rd ed.) Berlin: Springer-Verlag, T. (2001.

[16] H. Ritter, M. Thomas, & K. Schulten, Neural Computation and Self Organizing Maps. New York: Addison Wesley Pub. Co., 1992.

[17] D. O. Hebb, The Organization of Behavior: A Neurophysiological Theory, New York:Science editions 1949.

[18] A. Rosen A., & D. B. Rosen, The Application of the NCC to Sensory-motor Control of the Somatic Motor System. The Design of a Volitional, Obstacle Avoiding Multi-tasking Robot. Machine Consciousness Technical Journal. 1, 41-56 2003, (available for viewing at www.mcon.org)

[19] F. H. Guenther, D. Bullock, D. Greve, & S. Grossberg. Neural representations for Sensorimotor Control III. Learning a Body Centered representation of a Three Dimensional Target Position. J. of Cognitive Neural Science. 13, 341-358, 2001.

[20] S. Grossberg, The adaptive Brain II: Vision, Speech Language, and Motor Control, North Holland: Elsevier science Publisher, 1987.

[21] A. Rosen A., & D. B. Rosen, The Design of the NCC for visual perception: Solving the Inverse Optics Problem of “S

Forum Discussion