DARPA’s ACE Wants To Automate Dogfighting To Empower AI
A DARPA program to build pilot trust in artificial intelligence (AI) in combat by automating dogfighting has taken a step forward with the award of the first contract and the release of a solicitation to develop algorithms for close-in air combat.
Just as air combat training is used to develop pilot performance and trust, DARPA’s Air Combat Evolution (ACE) program aims to use dogfight automation as a crucible to give pilots confidence that AI “can handle a high-end fight.”
ACE will develop AI algorithms to automate within-visual-range combat and test them first in a modeling and simulation environment, then in subscale unmanned aircraft and finally in live 1v1, 2v1 and 2v2 combat between full-size fighters.
DARPA argues that dogfighting is less difficult to automate than it appears because the solution space is bounded. Instead the challenge is enabling the pilot to trust the AI, and ACE will develop methods of modeling and measuring trust, including crosscheck ratio—essentially how often the pilot grabs the stick to counteract the automation.
The first contract under ACE, worth $10.6 million, has been awarded to Lockheed Martin’s Skunk Works for Technical Area (TA) 3 of the program. This involves scaling the algorithms from local combat autonomy to large force exercises.
TA3 is one of four technical areas under ACE. TA1 involves developing algorithms for local combat autonomy; TA2 is building trust in those local behaviors; and TA4 involves integrating those algorithms into full-scale aircraft and conducting live flight experiments.
DARPA plans to awards contracts to two performers for TA3, scale-up, and one each for TA2 and TA4. The broad area announcement for TA1, algorithm development and demonstration, was released on March 6 with multiple awards planned.
In a bid to mobilize the AI community around the idea of automating within-visual-range combat, DARPA has sponsored the AlphaDogfight Trials. This is a virtual competition to demonstrate AI algorithms ahead of the TA1 development phase.
Eight teams from academia and industry were selected in October 2019. The final event at the AFWERX innovation hub in April will pit the best AI-based dogfighting system against the best pilots from the U.S. Air Force Weapons School, says DARPA.