Document Type


Publication Date



Air Force mission and training scheduling is an immensely complex, time-consuming, and significantly manual process. A scheduling tool known as Puckboard has been developed to help C-17 squadrons transition from moving pucks across large whiteboards to utilizing technology to dynamically plan and deconflict resources in the presence of complex constraints. The overarching goal of incorporating artificial intelligence (AI) into this tool is to empower schedulers to quickly produce more efficient schedules that promote unit readiness, with more pilots completing their training syllabi faster, and with fewer disruptions to missions, training, and aircrew personal life. Our AI efforts focused on refining a neural network approach combining reinforcement learning with linear programming to generate optimal schedules across varying timeframes. The development of this AI-enabled pilot scheduling tool involved applying human-centered design best practices, namely actively involving end-users to inform persona generation, tool functionality, existing and AI-enabled workflows, and wireframe development and iteration.