SBIR-STTR Award

The nature of tactical Army warfare is moving toward greater complexity as the command structure is required to deal with increasing amounts of information in near real time that must be used as a basis for critical decision making. The technology being developed in the artifical intelligence community is showing considerable progress to provide the techniques to help the battlefield commander meet this challenge. Processing areas include expert systems and neural networks. Phase I demonstrates application of techniques derived form natural selection and genetics (i.e., the genetic algorithm) to evolve finite state machines that learn to recognize patterns and compare results with a neural network approach. Finite state machines are a very simple way to transform digital data by mapping the observed data into predicted data by transitions between a finite number of internal states. The internal operation of the finite state machine is carried out in such a way as to characterize information content. Their simple structure along with the genetic algorithm is amenable to being implemented in efficient parallel processing architectures.

Benefits:
Application of the technology could be in situations where rapid processing and interpretation of data is critical. Examples are in robotics, nonlinear signal processing and control system design

As tactical warfare continues to accelerate with ever-greater complexity, battlefield commanders must increasingly deal with larger and larger data bases generated under dynamic situations. Typically, this data solicits time critical decision having far reaching consequences. This situation results from the unpredictable nature of modern warfare, and the tremendous increase in communications, computational, speed and memory capabilities. Today's technical community now face serious challenges to exploit new and evolving technologies. Digital networks appear promising as a decision aid in such areas as pattern recognition and classification, signal processing and robotics. One specific requirement is to develop new techniques for terrain analysis. Phase I demonstrated that a genetic algorithm could be used to train DPLM networks to perform low-level reasoning to recognize potential approach avenues for standard terrain data bases like the Tactical Terrain Data (TTD) base. This was then regulated by a payoff function which defined the total network performance. Phase II proposes development of a reasoning system for determining approach avenues by combining the low level feature recognition and using the genetic algorithm to train digital networks with a higher level reasoning concept. Tasks include development of a hardware system to operate on the TTD for demonstration.