DARPA has a collaborative air combat autonomy program that has made significant strides in recent months. The program is called Air Combat Evolution (ACE), and DARPA has announced that it’s halfway through phase 1. During its progress so far, ACE has eclipsed several key accomplishments in anticipation of live subscale aerial dogfights in phase 2 later this year.
Thus far, among the achievements made by ACE are virtual AI dogfights that involve within visual range and beyond visual range, multi-aircraft scenarios using updated simulated weapons. ACE has also performed live flights of an instrumented jet to measure pilot physiology and trust in AI. DARPA also notes that initial modifications to the first full-scale jet trainer scheduled to host the onboard AI pilot in phase 3 of the program have begun.
The biggest focus at the end of phase 1 is on the simulation-to-real transition of the AI algorithms as the researchers prepare for live-fly aircraft scenarios in late 2021. The transition to the real world is a critical test for most AI algorithms. According to the program manager Colonel Dan “Animal” Javorsek, prior efforts to transition AI to the real world have shown that some AI solutions can be overreliant on the digital artifact in the simulation environment.
When it began last year, the goal of the ACE program was to develop a trusted, scalable, human-level AI-driven autonomy for air combat using human-machine collaborative dogfighting as a challenge problem. In February, ACE algorithm development teams completed the next level of simulated AI dogfights in Scrimmage 1 at the John Hopkins Applied Physics Laboratory.
The team demonstrated 2-v-1 simulated engagements with two friendly blue F-16s fighting as a team against an enemy red aircraft. The contest was the first AI scrimmage that followed the AlphaDogfight Trials introducing more weapons, including a gun and a missile. Another major goal for the ACE program is to measure pilot trust in the AI ability to conduct combat maneuvers with the human pilot focusing on higher-cognitive battle manager decisions. Phase 3 of the program is currently set for late 2023 and 2024 using an L-39 full-scale jet trainer with the AI embedded.