Novel Fighter Combat Maneuvers Using Genetic Algorithms and Machine Learning

Technological advances have altered the nature of air warfare. In the past, air combat was frequently examined in the context of large scale air battles governed by attrition rates and quantity-quality trade-offs. Modern air combat has changed with the introduction of technologies such as stealth, advanced avionics, high off-boresight missiles, and helmet displays. Adequate exploration of combat maneuver possibilities requires some form of automated search and optimization process. Off-line methods are needed to define and evaluate optimal combat tactics efficiently and systematically. This, the area of tactics optimization is fertile for the application of genetic algorithms and machine learning approaches. This Phase I project will demonstrate the feasibility and applicability of tactics and maneuver optimization using genetic algorithms (GAs) and machine learning. This project team has developed pioneering state-of-the-art algorithms in the areas of differential game theory, Markov decision problems, machine learning), genetic algorithms and classifier systems, reinforcement learning, and min-max problems. Specific Phase I tasks are: 1) Acquire or develop a simulation of one on one air combat engagement. 2) Develop novel fighter combat maneuvers using genetic algorithms and machine learning. 3) Test and evaluate new fighter combat maneuvers on advanced one on one air engagement scenario. 4) Submit final report and Phase II recommendations Phase II will extend the air to air engagement to a M vs N model, specifically the MIL-AASPEM II model. Tehcnical support will be provided by leading experts from the fields of GAs and classifier systems (Goldberg,Smith) and differential games (Basar). McDonnell Douglas Corporation will provide technical and commercialization support during all phases of the project.