Eight teams have been selected by the Defense Advanced Research Projects Agency to participate in a virtual competition that seeks to demonstrate the use of artificial intelligence algorithms in simulated aerial combat.
Participants will utilize the open-source JSBSim flight dynamics model and the FlightGear simulator to compete in three challenges over the course of AlphaDogfight Trials, the agency said Monday.
DARPA introduced the AlphaDogfight Trials with the intent to assemble a network of AI developers as part of the Air Combat Evolution program, the agency’s initiative to automate aerial combat and address human trust issues with the technology.
The first two events are scheduled to occur next month and in January at Johns Hopkins University Applied Physics Laboratory. Each team will aim to counter agency-provided adversarial AIs through a one-on-one dogfight series. DARPA noted it will increase the level of difficulty in the second trial.
The third trial is scheduled to happen in March at the Air Force™s AFWERX innovation hub near Las Vegas, followed by the final matchup at Nellis AF Base. The agency plans to downselect two algorithms for the last demonstration event where U.S. Air Force pilots and contractor teams will be present.
“In the larger ACE program we want to demonstrate that human pilots teamed with AI can achieve greater effects in aerial combat than either could achieve alone,” said Lt. Col. Dan Javorsek, ACE program manager in the DARPA Strategic Technology Office.
The competitors are:
- Aurora Flight Sciences, a subsidiary of Boeing (NYSE: BA)
- EpiSci
- Georgia Tech Research Institute
- Heron Systems
- Lockheed Martin (NYSE: LMT)
- Perspecta (NYSE: PRSP)
- physicsAI
- SoarTech
