Intelligent Course of Action (ICOA) Generation for Air Vehicle Self-Defense

ABSTRACT: We propose to research a real-time tool for Planning of Intelligent Course of Action with Learning (PICAL). PICAL will reason over the available resources, available maneuvers, current threats and potential future threats and will generate a course of actions to maximize the chances of mission success. This course of actions can then be presented to the pilot for approval and modifications if necessary. The challenges in building PICAL robustly are: (a) to be able to plan in real-time while reasoning over a large number of relevant factors and a large set of possible actions; and (b) to be able to predict in advance what resources (e.g., ammunition) the remaining portion of the mission will require. To address the first challenge, we build upon our recently developed Anytime D* planning algorithm the first version of A* graph search that is both anytime and incremental. We will also build upon our recently developed concept of time-bounded graph construction designed specifically for planning with a large number of relevant factors in real-time. To address the second challenge, we propose to learn the desired level of remaining resources as a function of the mission type and environment and use it in the optimization process. BENEFIT: There are many reasons why a pilot cannot be expected to optimize manually in real-time the use of available weapons, counter measures and possible maneuvers in order to maximize the chances of mission success. First, people are generally not very good in searching for close-to-optimal strategies. Second, it is even harder to perform this optimization in real-time, and especially under attack. Third, people are not very good at estimating and manipulating the uncertainties which are inherent in missions that involve imperfect knowledge of threats and their capabilities. Unlike humans, computers are much better in dealing with all of these challenges. They are used abundantly to search for optimal and close-to-optimal solutions. Their performance is independent of whether the pilot is distracted. And finally, they are much better in numeric optimization under uncertainty. We therefore propose to build a real-time tool for Planning of Intelligent Course of Action with Learning (PICAL). For bomber pilots, PICAL will serve to conserve countermeasures to ensure that unique capability is not exhausted when alternative and more plentiful tactics could be used against threats. For general aviation pilots, PICAL will serve as training and/or a game. PICAL and its variants will serve the growing demand for Apps in the robotics, ipad, gaming, simulation, and artificial intelligence markets. For unmanned aircraft systems, PICAL already serves as the software engine for DPI"s family of highly autonomous unmanned aircraft systems for multiple military and commercial users.