Using Neural Networks for Battlefield Landmark and Feature Recognition

Reticular Systems proposes to research and develop a system for the recognition of landmarks and locations from image data, with emphasis on its use as a component in an autonomous navigation system for robotic ground vehicles. the mthodology proposed is based on a hybrid system consisting of both model-based object recognition techniques and neural network classification of terrain/landmark types. The information from these two sources is combined by an integration and inferencing component which performs high level reasoning to aid in location recognition and serves as a consistent interface to route planning and position estimation systems. We will deliver prototype code, simulation results and a final report describing the overall landmark and location recognition system for autonomous navigation of robotic ground vehicles. The report will detail information requirements, formats, data structures, algorithms, and initial results obtained in the prototype implementation or planned for Phase II. Prototypes and neural netowrk simulators will be used to evaluate the key components of the propsed system and reduce risk for Phase II. Phase I schedule and scope is realistic and realizable due to our experience in knowledge acquisition, intelligent systems, software development and development of prototyping tools.