Intelligent Course of Action Learning System (iCOALS)
ABSTRACT: SSCI proposes to create an Intelligent Course of Action Learning System (iCOALS), building on extensive past experience in learning advanced fighter combat maneuvers from simulation. By learning from repeated simulation, robust strategies can be constructed that allocate a complete action repertoire (including the mission-specific, limited range of weapons, countermeasures, associated maneuvers, etc.). By utilizing SSCI"s unique genetics-based machine learning techniques, these strategies take the form of rule sets, which can then be used for real-time reaction to evolving threats during mission execution. By evolving against a range of simulated mission conditions, the resulting rule sets will be robust against a corresponding range of IADS behaviors. This robustness can be further advanced by two-sided learning in the simulation, where the IADS also evolves strategy rule sets. iCOALS will encode strategies as rule sets that are highly human-readable. This will enhance the ability for human intervention in a semi-autonomous execution mode. iCOALS will also draw on recent advances in the genetics-based machine learning strategies utilized in the original fighter combat. These advances draw on information theory, and have demonstrated world-best performance on complex learning problems, while decreasing rule set size, improving accuracy, and improving human readability of resulting rule sets. BENEFIT: iCOALS can employ faster-than-real-time offline simulations with perturbations and realistic constraints to formulate a package of rules and parameters that react in real time to varying threats. Numerous threat tracks can be used in simulation to form a robust strategy that will react to different tracks in real time.The iCOALS action repertoire will include all these, and the simulation basis will insure they are evaluated in a vehicle-realizable context, including sequencing. A pre-planned route will form the fundamental constraint in the simulation from which iCOALS learns. Maneuvers must explicitly conform to this, and all other vehicle constraints. Allocation of limited resources is implicit in the learning of iCOALS strategies. By introducing different mission parameters and models into the iCOALS simulation, the system immediately adapts a strategy to that particular mission. iCOALS directly provides a"strength"for each rule, for real-time indication of threat priority. iCOALS basis in simulation and heuristic learning means that extensive computational models are not necessary. The human-understandability of iCOALS rule sets will facilitate user interaction for semi-automatic plan execution.
Small Business Information at Submission:
Sr Group Lead: Intelligence Network
Scientific Systems Company, Inc
500 West Cummings Park - Ste 3000 Woburn, MA -
Number of Employees: