Genetic design of processing elements for path planning networks

The genetic algorithm (GA) is applied to the design of processing elements (PEs) for recurrently connected optimal-path-finding networks. A PE is represented by a parameterized functional expression, and the genetic algorithm searches the function's parameter space to optimize network performance. The GA evaluates each parameterized processing element by placing the PE in a recurrent network, initializing the network with data from an actual terrain, and comparing the relaxed network state with one incorporating the known optimal PE function. For the network topology considered, it was found that a composition of sigmoid terms was able to approximate a discontinuity which existed in the solution PE function. The genetic algorithm employs the principles of the adaptive representation genetic optimizer technique (ARGOT), which increases search speed and efficacy. The present investigation suggests an approach to ANS development which treats the PE function as a design variable