Planning the delivery of cargo as part of an airlift operation is a notoriously complex problem. Transportation routes can suddenly become inaccessible due to poor weather or other unexpected occurrences. Factors such as airplane speed, carrying capacity, and airport maximum-on-ground must also be considered to ensure on-time and efficient delivery. Unforeseen cargo needs may require quick re-planning to meet tight deadlines.
A potential solution may be found at the AFRL Airlift Challenge, an online multi-agent planning competition where competitors create innovative algorithms and solutions to execute a simulated airlift operation.
The competition uses artificial intelligence, or AI, to assist in the design of plans for the aircraft to follow to efficiently deliver cargo. For each airport at which an aircraft stops, the AI provides an "order" designating which cargo to load or unload, as well as the next destination.
"A large demand and tight deadlines make airlift operations difficult to plan even under ideal conditions," said Dr. Andre Beckus, AFRL machine learning researcher. "Unexpected disruptions only further complicate the problem, potentially introducing major delays and stressing planning software to its limits."
To identify new solutions, researchers apply their skills and develop new algorithms that can achieve on-time deliveries while improving efficiencies. A small example scenario is shown in the image.
The AFRL AI Airlift Challenge scenario shows how AI can deliver cargo. Airports are identified with small squares with connecting routes (white lines). Cargo is staged at three airports in the pickup area (green rectangle). Each is designated for delivery at a specific airport in the area of need (yellow circle). The developed algorithm guides four airplanes through the network to pick up and deliver the cargo. The routes undergo random disruptions, requiring aircraft to either wait for the disruption to clear, or follow a different route.
""The Airlift Challenge provides a simulation environment in which artificial intelligence, or AI, agents can interact," Beckus said. "The algorithms were scored against a set of increasingly complex evaluation scenarios while contending with unexpected events and disruptions."
The information gathered from this AI challenge and others, such as the AFWERX Expedient Basing Open Innovation Challenge, can assist the USAF in successfully innovating new ways to accomplish traditional mission objectives of air lift operations more efficiently and effectively on small or even larger scale, like the historic Berlin Airlift in 1948.
During the competition, participants submitted AI agents for immediate scoring to see their rank on a real-time leaderboard.
"It was exciting to see researchers around the world be able to access this environment and provide solutions for the given scenarios" said Adis Delanovic, computer scientist at AFRL. "We look forward to holding more competitions such as this one where we can use the untapped potential of crowd sourcing solutions to warfighter needs."
AFRL researchers and the winners of the competition had the opportunity to publish a joint paper on the results of the AI Airlift Challenge at the SPIE Defense + Commercial Sensing conference April 30 through May 4, 2023, in Orlando, Florida.
The winner of the Airlift Challenge, Dr. John Kolen, is an independent researcher from Florida. Dr. Kolen has over 30 years of experience in the industry. The second-place runner up was a team of researchers from the Raytheon Technologies research center, which included Dr. Abeynaya Gnanasekaran, Dr. Amit Surana, Dr. Kunal Srivastava, Dr. Hongyu (Alice) Zhu, and Dr. Yiqing Lin. An honorable mention went to students from the University of Florida which included Nickolas Arustamyan, Norman Bukingolts, Dali Grimaux-De camps, Matthew Huynh, Adam Sardouk, and Devin Willis. Their advisors were David Bragg; Florida Applied Research in Engineering (FLARE) and Dr. Kaleb Smith; Senior Data Scientist, NVIDIA AI Tech Center.
These participants altogether uploaded 92 submissions in which they iteratively refined their algorithms. "We were impressed with the high quality of the solutions submitted by the winners," said Beckus. "The teams were able to quickly improve their scores in a short amount of time."
The challenge received interest from 27 registered users on the competition platform, as well as 150 followers on the challenge.gov competition hub. The competition also generated valuable discussion and feedback from participants and allowed the government team to improve the simulator.
To learn more and to register for the next competition, visit https://airliftchallenge.com.
Related Links
Air Force Research Laboratory
Aerospace News at SpaceMart.com
| Subscribe Free To Our Daily Newsletters |
| Subscribe Free To Our Daily Newsletters |
