General Atomics Aeronautical Systems (GA-ASI) has conducted a flight test to further validate the firm’s Collaborative Combat Aircraft system in El Mirage, California.
The demonstration saw an MQ-20 Avenger unmanned aircraft take off from GA-ASI’s Desert Horizon facility for three autonomous combat simulations.
The aircraft was paired with an MQ-20 “digital twin” to assess the CCA’s artificial intelligence and machine learning capability in live, virtual, and constructive multi-objective missions.
Deep Learning in Flight
The aircraft employed GA-ASI’s reinforcement learning (RL) architecture to develop and apply deep learning algorithms for various actions.
During the demonstration, the single RL agent assisted the live plane in avoiding threats while the multi-RL agent flew both the live and virtual MQ-20s as they deflected and chased targets.
Meanwhile, the hierarchical RL agent leveraged sensor information to choose the drone’s next steps depending on the scenario.
Evolving AI-Based Drone Pilots
According to GA-ASI, all three RL agent models proved the Collaborative Combat Aircraft’s (CCA) pilot capability to complete aerial missions by exploiting real-time information and processing critical decisions without the help of human operators.
“The concepts demonstrated by these flights set the standard for operationally relevant mission systems capabilities on CCA platforms,” GA-ASI Advanced Programs Senior Director Michael Atwood stated.
“The combination of airborne high-performance computing, sensor fusion, human-machine teaming, and AI pilots making decisions at the speed of relevance shows how quickly GA-ASI’s capabilities are maturing as we move to operationalize autonomy for CCAs.”
