CAMBRIDGE, Mass. -- The Department of the Air Force Artificial Intelligence Accelerator Program developed a new tool, the Puckboard Intelligent Recommendation Engine, which has the potential to improve mission readiness through smarter, more efficient aircrew mission and training scheduling.
Aided by modern web and algorithm technologies, the DAF-AIA delivered a collection of novel methods for building intelligent mobility aircrew schedule plans that maximize training and readiness while meeting real-world mission demands. These methods, including the AI-planning optimization approach in operational usage, and research into deep learning-based planning, give Mobility Air Forces squadrons a major step towards exiting a historically manual and rules-based approach to schedule planning.
Scheduling aircrew to operate an aircraft for sorties requires numerous considerations, to include if each Airman has the necessary training completed, who needs additional flight hours, and even scheduled leave/paid time off.
“The problem of scheduling flight-hours has long been complex, time-consuming, and manual,” said Col. Garry ‘Pink’ Floyd, DAF-AIA director. “Our team developed an AI-enabled mission and training scheduling system that we think will boost readiness through the use of smarter tools that bring aircrew members and schedulers together toward a common goal: making sure our aircrews are ready. The goal is to boost readiness, not necessarily by flying more sorties, but by flying smarter schedules.”
“There’s a complex web of constraints that must be modeled in this problem, from crew legality, through to training needs, qualification levels, crew rest, and more, all with aircrew members who must balance obligations to their administrative and operational chain of commands,” said Maj. Eric Robinson, project lead. “Taking these considerations into account, along with building quantifiable mechanisms for defining ‘good’ schedules, the team built an intelligent recommendation system that has been researched, prototyped, and rigorously tested prior to delivery to the field.”
The development team used data from approximately six months’ worth of training and mission flights at Charleston Air Force Base. This data collection was sponsored by the Chief of Data and Artificial Intelligence Office and AIA through public ‘datathons’ in 2020 and 2021. The datathons used MIT research that combined mathematical optimization with a type of machine learning called reinforcement learning, and applied it to a specific mobility use-case (learn more about the fundamental research here: NICE: Robust Scheduling through Reinforcement Learning-Guided Integer Programming | Proceedings of the AAAI Conference on Artificial Intelligence).
The team continued to iterate.
"We know that we increase our operational readiness and lethality per gallon by flying smarter, not less. Credit must be given to the innovative minds of the AI Accelerator team for spearheading the development of AI applications within Puckboard that will allow us to do just that.” said Mr. Roberto I. Guerrero, Deputy Assistant Secretary of the Air Force for Operational Energy, Office of the Assistant Secretary of the Air Force for Energy, Installations, and Environment. “When we initially saw Tron’s Puckboard application in late 2019, we knew this was an ideal initiative to invest our funds in. We are eager to support these types of transformative tools that will help us fuel more fight."
The Office of the Assistant Secretary of the Air Force for Energy, Installations, and Environment funded the first production version of intelligent scheduling recommendations in late 2022. This version used a variation of AI-planning optimization techniques developed during research that was designed to blend with machine learning approaches like NICE in the future, while maintaining the explainability and performance of constraint-based techniques.
Then, working with Air Mobility Command leadership, the team achieved an authority to operate through DoD Platform One to test mission scheduling at operational squadrons. They worked to connect the production intelligent recommendation engine to the Puckboard product and user interface throughout 2023. The first successful human-machine workflow recommendations were produced in September 2023.
“There is still work to be done, but the fact that we have real research, implemented on a real use-case, with real end-users, providing real feedback is the biggest tangible step we have seen in 75 years towards modernizing our scheduling processes,” Robinson shared. “Seeing our teams at the AIA, MIT, AMC, and PEO Digital research, prototype, and implement this solution from a blank slate shows the power of bringing our operational experts together with the world’s top researchers to solve national security problems.”
Currently the team is continuing to advance the product with beta users at Travis and McChord Air Force Bases. Additionally, the AIA and AMC officially completed the Program Management transition of the intelligent scheduling recommendation engine and Puckboard products into the PEO Digital portfolio in February 2024, retaining AMC as the lead command. PEO Digital will continue to advance and scale this work in the coming years.
Colonel Floyd believes the AI Accelerator's experience with building intelligent schedules for Puckboard provides a powerful example of how the DAF cooperates with academia and industry.
“I could not be more proud of this effort. We connected fundamental research with an operational need, and now Airmen at four bases have the opportunity to evaluate the prototype and provide feedback,” Floyd stated. ”As the intelligent recommendations improve and scale to other platforms, it is entirely possible that they will play a major role in boosting readiness. There is a real chance we could see a 5% or 10% boost in readiness from the use of this tool. We’ll see what happens, but if we land even close to that, the AI Accelerator's Intelligent Scheduling team will have had a huge impact on operations for years to come.”