AI/ML For Kinetic and Non-Kinetic Missile Defense Battle Management

The modern battlefield has seen an increase in the diversity of threats, resulting in increased complexity for Combatant Commanders determining weapon selection and fire control. These complex decisions intensify under the duress of multi-threat raid scenarios and motivate the creation of an automated real-time tool to aid in weapon-target pairing decisions. AI/ML algorithms are prime candidates for this weapon selection tool, but the diversity of AI/ML algorithms, ranging from simplistic decision trees to complex reinforcement learning, warrant a trade study to identify the benefits and costs of each potential approach. Additionally, AI/ML models require large amounts of data to be effective, and a successful framework should also identify data sources for training. This proposal seeks to address both the AI/ML diversity and the dataset scarcity to develop a holistic approach to automated weapon selection recommendation. AI/ML approaches such as supervised learning, unsupervised learning, and reinforcement learning will be investigated and applied to the various stages of weapon selection, including determining weapon efficacy, maximizing probability of engagement success criteria, and incorporating additional penalty functions such as weapon cost and weapon magazine depletion. Parallel to this trade study, the lack of large, available datasets will be mitigated through the development of a prototype simulation suite capable of synthetic weapon selection decisions based on incoming threats and environmental factors. This simulation suite will enable database generation of parametrically generated scenarios utilizing kinetic and non-kinetic weapon systems, laying the initial groundwork for training and testing candidate AI/ML models. Approved for Public Release | 24-MDA-11906 (16 Sep 24)