Tactical AI for Marine Corps

Planning and Re-planning: Decision aids to increase the speed and accuracy of the planning/re-planning cycle.

• Task Generation: Recommend tasks for a particular mission (commander’s intent) to maximize outcomes.
• Branch/Sequel recommendations: Recommend contingency options based on predicted potential mission tasking, resourcing, or other mission state changes. Provide follow-on tasking based on predicted mission end states. Of particular interest are real-time branch/sequel recommendations in support of dynamic re-planning, using execution time mission state parameters as input. 
• Force Allocation and Composition: Model optimal force composition and allocations based on required tasks, weapon/target pairing, enemy force composition, forces readiness, and available resources.
• Opportunity Recognition: Identify possible decisive points relevant to a mission to provide Blue forces with opportunities to simulate multiple COAs, wargame or re-plan.
• Enemy Course of Action: Predict current enemy course of action as well as reactive ECOAs based on various blue force actions.  
• Automated, geo-specific 3D terrain generation from a variety of source data to support organic content generation for mission rehearsal and planning. 

Mission Execution Monitoring: Model critical mission tasks and monitor those tasks as they occur during a mission to provide actionable recommendation or alerts to the warfighter.

• Measures of Effectiveness / Measures of Performance: Estimate MOEs and MOPs for the current mission during execution and predict end-state MOEs/MOPs. Provide higher level assessments that incorporate multiple MOEs and MOPs into assessments that are easy to understand.
• Critical Information Requirement Alerting: Identify and model critical Priority Intelligence Requirements (PIRs) and Friendly Force Information Requirement (FFIRs) in support of alerting the warfighter when they are triggered or predicting their occurrence later in the mission.  Recommend subsequent actions to maximize mission performance based on the PIR/FFIR triggers.
• Mission Relevant Information Management and Extraction: Develop methods of managing multi-modal mission relevant information in support of maximizing its utility to machine learning and artificial intelligence modeling techniques, especially in tactical environments. Develop services to extract relevant data in response to current or predicted mission events. 
• Platform Task Recommendations: Automate or recommend tasking of unmanned and manned platforms, as well as small units based on mission objectives. Tasking models should continuously monitor the execution of the mission and adjust recommendations based on the changing mission state.
• Tactical ATR: On-board automated target recognition for UAVs or UUVs to support identification of targets and communication of those identifications over bandwidth constrained communications networks. There is interest in identification of humans, equipment, and weapons across multiple sensor modalities (EO, IR, SAR, etc.) and in multiple environments (air, water, ground).

Human-machine Teaming: Maximize performance of warfighter interaction with artificial intelligence based decision support tools.

• Human Computer Interface: Design and develop interfaces which effectively communicate decision aid results to the warfighter without incurring cognitive overload. Develop interfaces which work effectively in a tactical environment.
• AI Trust: Research the factors which increase or decrease warfighter trust in AI-based decision tools. Develop new techniques to convey decision tool uncertainty to the warfighter such that trust is properly aligned to accuracy of results.

• Data Feature Engineering: Ensuring all data relevant to a question is utilized with realistic quality levels

• Developing AI techniques that can be used effectively in resource constrained tactical environments.

• Defining Reusable algorithms and realistic algorithm training data to rapidly extend mission capabilities.

• Fusing of attended and unattended model results: Optimize model performance by leveraging data that spans a large feature space and that both contains and supports human insights.

• Developing dynamic explanatory tools that efficiently presents the back story of how and why an algorithm recommended what it recommended