Continuous Interactive Learners for Mission Planning (CILEMP)

A key problem with automated mission planning software for military commanders is that their underlying domain models may be incomplete, imprecise, inaccurate, or inefficient and may rapidly become obsolete because of evolving domains. These issues can adversely impact the quality and the effectiveness of support offered by automated software. However, manual maintenance of models planning domain is slow, expensive and error prone. Recent advances in machine learning (ML) present an opportunity to address this problem by automatically acquiring, enriching and updating asset performance models and tactical planning knowledge for use by automated mission software. We will develop CILEMP, a software system comprising Continuous Interactive Learners for Mission Planning. CILEMP shall utilize a family of numeric and symbolic learning approaches to learn from subject matter experts(SMEs) inputs they provide during practice planning sessions and during real operations. We will investigate ML approaches such as learning from demonstration and learning from experience to acquire and update model parameters, planning preferences, and higher order structures for tactics to improve model coverage, precision, accuracy and efficiency. We will work with SMEs to develop scenarios for evaluating the feasibility of CILEMP. We will develop a software architecture specification and prototype implementation plan for Phase II.