AlphaICE: Autonomous Intelligent Combat Engagement

EpiSci proposes AlphaICE: Autonomous Intelligent Combat Engagement, a hybrid & adaptive application of EpiSci’s Tactical AI multiagent autonomy framework to deliver dominant tactical autonomy across the full spectrum of fighter air combat mission sets. Existing approaches suffer from limitations in mission scope & scale, fragility in the presence of uncertainty or limited observability, and limited online adaptability to dynamic tactical environments. Recent AI developments, including highly scaled deep reinforcement learning (DRL) processes, offer promise but struggle with management of training scope as well sim-to-real (sim2real) transitions; specifically, sim2real predictivity where the delta in performance was due to exploitation of actions not available in live-flight environments. AlphaICE manages the complexity of all-domain air combat via an operationally informed, hybrid, hierarchical system design. Utilizing a novel hybrid approach, Tactical AI systems combine the exceptional performance and task adaption of modern AI techniques (such as DRL) with the assured, bounded behavior and rapid integration of traditional deterministic control systems and heuristic-based collaboration.Tactical AI is built via an Agile development process, capable of weekly (or faster) code deliveries for rapid iteration, frequent testing & operator feedback, and early identification and resolution of challenges and risks. The prior and parallel work performed by the EpiSci team will reduce the considerable technical and cost risks of the ambitious goals, maximizing the likelihood of success and the development of transitionable capability. The context in which AlphaICE will be developed and tested calls for a mission-oriented test methodology, with key tracking measures that reflect a focus on delivery of operational capability. A collection of measures derived from developmental and operational fighter aircraft test and evaluation is proposed. These measures will be used at every stage of development and test, providing a consistent evaluation framework primarily centered on mission effectiveness. These measures of effectiveness can be further decomposed into measures of performance and measures of suitability which address all phases and aspects of the mission, from mission planning to development and execution of kill chains. From this common set of metrics, the AlphaICE team will execute a data-driven test approach to meet the program goals of lowering disengagements, increasing the sim2real correlation coefficient, and shortening Mean Time to Recover from unexpected parameters and results while improving mission effectiveness and performance.