SCIENTIFIC SYSTEMS CO INC — Department of Defense SBIR Phase II: SSCI proposes a retooling of the Intelligent Course of Action Learning System (iCOALS) tha

Description

SSCI proposes a retooling of the Intelligent Course of Action Learning System (iCOALS) that leverages the important results from the Phase I effort and remedies the deficiencies of a rule-based approach when applied to complex Pursuit Evasion Games (PEGs). This modification is critical given the utility of the proposed approach to DARPA's Anti-Submarine Warfare (ASW) Continuous Trail Unmanned Vessel (ACTUV) program. ACTUV's"Track-and-Trail"problem is specific, real world instance of a PEG where ACTUV attempts to maintain close in sensor contact over a period of weeks with a manned diesel-electric submarine. This problem can be formulated as a PEG where the unmanned pursuit vehicle (ACTUV: propulsive superiority) attempts to maintain proximity with the evader (submarine: intellectual superiority). SSCI's specific approach to this problem provides an online Adversarial Autonomy (AA) engine that employs a Forward Reachability Model (FRM) formulation of a PEG game. This is important because it is impossible to formulate a rule-set that can handle all challenges from an intelligent adversary especially one with access to external effectors (surface traffic, active/passive decoys, weapons systems, boarding parties, etc.). Game-theoretic approaches, like FRM, model key performance parameters of the pursuer (ACTUV) as well as modeling or measuring those of the evader. Once modeled, efficient computational search algorithms (alpha-beta) are employed to ensure optimal outcomes for the current engagement configuration in real-time. It should be noted that engagement-level optimality is precluded in a rule-based approach with a finite rule set. Upgrading iCOALS with online adversarial models will eliminate the possibility track-and-trail will be easily broken by an intelligent adversary who quickly discovers and repeatedly exploits rule-set loopholes.