Revisiting the docking of the truck: MacGregor's spiking neurons using a Three-Neuron Controller (TNC)

Traditional Artificial Intelligence practitioners viewed intelligence as a high-level cognitive function based on symbolic computation. Recently, an alternative view has arisen. Intelligence is seen as adaptive behavior, in systems which contain the intelligence "onboard" and which are situated in "real-world" environments. Work in these "situated and embodied" systems is often based on neural information processing systems. Parallel Distributed Processing (PDP) systems are no longer dominant. Rather, more neuromorphic systems are studied, for example Beer's hexapod model Periplaneta Computatrix (1992). As a result of this trend, we have revisited the work of Sorkin and Alexander (1998). This work employed MacGregor's (1987) spiking neuron model coupled with a three-neuron controller, in solving the truck backer-upper problem. This is a standard toy control problem. The task is to back a truck, located anywhere in a square loading area usually 100 by 100, and with the truck at any angle to the loading dock, to the center of the loading dock (usually at Cartesian coordinate) (50, 100) with the rear of the truck at the dock and the truck at perpendicular to the dock. A Three-Neuron Controller is a two-layered artificial neural network (ANN) having two neurons on the lower or input level, and one neuron on the upper or output level.