The Engineering Design of a Neuronal Correlate of Consciousness (NCC) Circuit for the Sensory-Motor Control of a Robotic Arm

Introduction: The NCC- Relational Robotic Controller (RRC) control of a Robotic Arm

Stephen Grossberg and his colleagues at the Boston University Center for Adaptive Systems best treat the problem of sensory motor Control without the NCC-Consciousness Mechanism (Grossberg 1982, 1987, Guenther et al, 2001). The search for the equations of motions underlying inter-limb coordination and inter-limb dynamics has been described by Swinnen et al (1994). Many of the neural network computational techniques were developed by Teuvo Kohonen and Helge Ritter in the study and application of self-organizing maps to sensory-motor control Problems (Kohonen, 2001, Ritter et al, 1989). For example, in “Chapter 11: Visuomotor coordination of a Robotic Arm”, Ritter et al make use of an extension of Kohonen’s model for the visuomotor control of a robotic arm that includes learning of control signals (note1). The learning algorithms used in this paper are derived from the ones described by Ritter et al (1989).

“At any given moment our brain manages to control 244 different degrees of freedom that involve more than 600 different body muscles. In fact, dozens of different muscles routinely act simultaneously. For example, the muscles of each arm or leg control 30 degrees of Freedom, utilizing rather complicated muscle combinations. The complexity of this control becomes obvious as soon as one attempts to equip a robot arm with only a small fraction of human dexterity” (Ritter et. al., 1989).

A robotic body simulating the human body consists of over 80 joints and 140 degrees of freedom (note 2). Each joint is characterized by 1-3 degrees of freedom, and simulated by 1-3 motors located at each joint. The most important characteristic of the NCC-RRC motor control system is that as hierarchy of similar (nodal map modules) circuits may be trained to control and coordinate the motion of robotic parts relative to any or every joint in the robotic body. Thus to control the inter-limb dynamics of the total robotic body, more than 80 trained nodal map modules are required, and each nodal map module must control 1-3 degrees of freedom. The engineering solution presented in this paper applies to all 80 joints of the robotic body. Thus this paper presents a building block for the engineering design of a dynamically coordinated sensory-motor control system of the human somatic motor system.

The design of the NCC-RRC

The “self” nodal map module and the training path (programming) associated with it, is neural net based, whereas the sequence Stepper Module and the Task Selector Module are designed by microprocessor based sequential algorithmic programming. The neural net-based portion of the design is presented in this example. Microprocessor based programming is a direct application of the procedures described by Rosen (2003, MCTJ_1:41-56), and also published on the website www.mcon.org. This example presents the neural network portion of the design (programming). Thus the p-field and q-field data modules and the training path for the “self” nodal map module, shown in figure 1, is neural net based, whereas the remaining modules (task selector, sequence stepper, and Control signal output module) are programmed by microprocessor based sequential algorithmic programming.

Conclusion

The foregoing is a reverse engineered building path for the control of 2 joints of a robotic arm. The solution, however, is applicable to all the joints of a robotic human-like body. A robotic humanoid body may be controlled by a hierarchy of 80 RRCs, each RRC controlled by a nodal map module devoted to one of the 80 joint in the body, and the design of each joint given by the solution presented in this paper. The hierarchical solution consists of massive parallel processing that simultaneously controls up to 240 degrees of freedom (80 joints time 3 degrees of freedom per joint). The simultaneous “self” training of the array of RRCs (total hierarchy of 80 nodal map modules) immediately yields a solution to the equations of motion underlying biological inter-limb coordination and inter-joint dynamics. The incorporation of the sensory Neuronal Correlate of consciousness (NCC) into the equations also generates a complete solution to the design of a volitional, obstacle avoiding, multi-tasking robot,

Comparison of the RRC with the biological brain. The design of the RRC adheres to neuro-biologic and Darwinian evolutionary constraints. Therefore the input output functional characteristic of the RRC depends on the fidelity that was achieved in adhering to those constraints. The structural design of the RRC, however, is implemented with a hybrid approach of neural net-based programming and microprocessor based-programming. The greatest fidelity to the structural portion of the brain is achieved with the neural net-based programming. But even in this case neural nets, made up of electronic neurons, are a crude approximation of biological networks made up of biological neurons. In contrast to rigid electronically designed neurons, the brain is functionally highly redundant, it is equipped with a complex array of structurally diverse biological neurons, with variable threshold functions, different synaptic communication methods, and dendritic-synoptic learning, or programming-communication that is bio-chemically and structurally diverse, functionally efficient, and variable depending on the function of each neuronal network.

The question of: to what degree does the RRC simulate the operation of the biological brain? is discussed in note 3 And the question of: to what degree do the electronic neural interconnections simulate the interconnections in the biological brain?, is discussed in note 4.

Forum Discussion