Teams get a few more months to hone their AI algorithms for virtual aerial dogfight competition
Apr 27, 2020
DARPA’s AlphaDogfight Trials Final event, originally scheduled for April, has been rescheduled for the summer due to the COVID-19 pandemic. The third and final trial is now planned for Aug. 17-20, 2020, at AFWERX, the Air Force’s innovation hub, in Las Vegas, Nevada.
Eight teams were selected last year to compete in a virtual competition designed to demonstrate advanced artificial intelligence (AI) algorithms capable of performing simulated within-visual-range air combat maneuvering, colloquially known as a dogfight. The series of trials was designed to energize and expand a base of AI developers for DARPA’s Air Combat Evolution (ACE) program. ACE seeks to automate air-to-air combat and build human trust in AI as a step toward improved human-machine teaming.
“AlphaDogfight Trials performers have used this down time constructively to refine their algorithms and to prepare for the final event later this summer,” said Lt. Col. Dan “Animal” Javorsek, AlphaDogfight Trials and ACE program manager in DARPA’s Strategic Technology Office. “Trial 3 will be the first time the teams compete their AIs against each other publicly. The winning AI will then fly a simulated dogfight against an experienced fighter pilot to test the AI’s mettle and demonstrate how far these algorithms have come in only a few months. It typically takes nearly a decade for human pilots to build the experience necessary to perform at this level.”
The plan is for an F-16 pilot from the Air Force Weapons School at nearby Nellis AFB to fly in an F-16 simulator against the Top AI in a mock 1-v-1 battle. Win or lose, the competition will have advanced AI-driven tactics in combat maneuvering and provided a path to further development.
The first trial took place Nov. 19-21, 2019 at the Johns Hopkins University Applied Physics Laboratory (JHUAPL) in Laurel, Maryland. The following video provides a recap of Trial 1, where algorithms were still in early development, and provides an overview of AlphaDogfight Trials:
Just a few months later, on Jan. 28-30, 2020, teams returned to JHUAPL for Trial 2, where they demonstrated greatly improved algorithm performance against a more capable adversary “red air” agent developed by JHUAPL. To better align with the ACE program and demonstrate the extensibility of the approach, the platform was also changed from the F-15 in Trial 1 to the F-16 in Trial 2. This video describes the AI improvements during Trial 2:
The following eight teams are preparing for participation in the Finals:
- Aurora Flight Sciences
- EpiSys Science
- Georgia Tech Research Institute
- Heron Systems
- Lockheed Martin
- Perspecta Labs
- physicsAI
- SoarTech
Should the ongoing COVID-19 pandemic crisis preclude traveling to AFWERX in August, the back-up plan is to conduct Trial 3 as a distributed competition. JHUAPL would coordinate the final event virtually, and teams would compete their AIs and conduct the final match-up between the Top AI and fighter pilot remotely.
# # #
Media with inquiries should contact DARPA Public Affairs at outreach@darpa.mil
Associated images posted on www.darpa.mil and video posted at www.youtube.com/darpatv may be reused according to the terms of the DARPA User Agreement, available here: http://go.usa.gov/cuTXR.
Tweet @darpa