| The successfulness of biological systems have encouraged the investigation and development of some of these biological principles into engineering applications. This area of development has been termed artificial neural networks. These artificial neural networks have been very successful for a variety of static mapping problems, for example character recognition, and optimization problems. However artificial neural networks utilized in dynamic problems have had less success, an example of this is the control of a robotic limb, where the more conventional control schemes are preferred over artificial neural network control schemes.;The high functionality of human limbs suggests that there may be some artificial neural network scheme that would provide superior control for robotic machinery. However many of the temporal properties present in biological neurons and networks have been ignored in artificial neural networks. The aim of this research is to develop a more biologically based artificial neural network, that will incorporate some of the temporal properties that have been excluded from present artificial neural networks.;A model was developed that relied on a frequency based internal signal. This facilitated the addition of several biological neural properties, such as synaptic fatigue, relative and absolute refractory period and capacitance neural membrane decays. These new properties incorporated more temporal information into the artificial neural network. The model was then able to train to single and multiple desired output frequencies, using an input mainly as the initiation signal.;The following step was to incorporate a control scheme that would manipulate an object along a spatial-temporal path. Since central pattern generators (CPG) are believed to be utilized by biological systems for the control of repeated movement, it was chosen to implement this type of control. A simple movement device, termed the roller racer, was utilized to implement the new artificial neural network and investigate the CPG control scheme.;The model was successfully able to train the roller racer to move along two different desired paths at three different rates. The results were sufficiently successful to demonstrate that some of the complex biological features should be included in artificial neural network for certain applications, and that CPG control schemes are an important method to control repeated moving systems. |
