SCIENTIFIC SYSTEMS CO INC — Department of Defense SBIR Phase I: MDA21-021

Description

Reinforcement Learning for Intelligent Salvo Management (RLISM) is an automated Artificial Intelligence (AI)/Machine Learning (ML) driven battle management software decision aid for intercept planning of ballistic missiles across multiple salvos. The decision aid system provides functionality for an operator-configurable offline (pre-mission) Monte-Carlo simulation across huge numbers of scenario permutations, in order to train the reinforcement learning (RL) algorithm components. RLISM provides an operator-configurable simulation and user interface to model, test, assess, and train operators with the RLISM online capabilities. Incorporated within RLISM is a pre-launch constraint satisfaction combinatorial solver, and associated user interface, to optimize, plan, and assess interceptor salvo compositions, configurations, and schedule timing. To incorporate in-flight updates, a mission management capability, and associated user interface, is provided for reactive engagement re-tasking in real time. The algorithmic basis to accomplish the above will combine algorithms for combinatorial optimization and reinforcement learning, to leverage the “best of both worlds” – the deterministic constraint handling of combinatorial optimization, with the speed and flexibility of RL. Approved for Public Release | 21-MDA-11013 (19 Nov 21)