Optimal path analysis using a predator-prey neural network model

A neural network research effort is currently underway at Rome Air Development Center, the Intelligence and Reconnaissance Division (RADC/IR). Griffiss Air Force Base. The purpose of this research is to solve computationally difficult intelligence exploitation problems that have eluded conventional techniques, e.g., target recognition, battlefield multi-sensor correlation and fusion, and intelligence situation assessment. This paper describes the use of a predator-prey neural network paradigm for path analysis. A proof-of-concept simulation is developed and successfully utilized to map optimal/near-optimal paths from given starting points to given destinations through a field of obstacles. The worst-case computational complexity for this algorithm, when implemented on a parallel architecture, is in order of &Ogr;(n), where n is equal to the number of nodes in the network. Serial implementations are in order of &Ogr;(n1.5). This is noteworthy because fast and reasonable solutions to complex problems are often preferable to an ideal optimal solution that typically requires specialized hardware and/or too much time and money to generate. Potential applications for this model include trafficability analysis and route prediction. This model could also serve as a pre-search tool to set search bounds for heuristic search algorithms such as A*. Application of this paradigm to the Enhanced Terrain Perspective Viewer is also discussed.