General Atomics Aeronautical Systems, Inc. (GA-ASI), an American manufacturer of unmanned aerial vehicles (UAVs), progressed Collaborative Combat Aircraft (CCA) ecosystem by flying three distinct missions with artificially intelligent (AI) pilots on an operationally pertinent Open Mission System (OMS) software stack. An Avenger® Unmanned Aircraft System (UAS) owned by the business was connected with “digital twin” aircraft to autonomously execute Live, Virtual, and Constructive (LVC) multi-objective collaborative combat missions. The flights, which occurred on December 14, 2022, from GA-Desert ASI’s Horizons flight operations facility in El Mirage, California, demonstrate the company’s dedication to evolving its CCA ecosystem for Autonomous Collaborative Platform (ACP) UAS using AI and Machine Learning (ML). This gives a novel and unique decision-making tool for military platforms of the future generation operating in dynamic and uncertain real-world situations.
The flight utilised GA-innovative ASI’s Reinforcement Learning (RL) architecture developed with agile software development methods and industry-standard tools, including Docker and Kubernetes, to build and evaluate three deep learning RL algorithms in an operationally realistic environment. Single, multiple, and hierarchical agent behaviours were exhibited by RL agents. The single-agent RL model navigated the live plane successfully while dynamically avoiding dangers to complete its objective. Multi-agent RL models flew a real and virtual Avenger to pursue and avoid threats while pursuing a goal jointly. The hierarchical RL agent utilised sensor data to select action courses based on its comprehension of the world state. This exemplified the AI pilot’s ability to independently assess and act on live, real-time data to make mission-critical choices at the speed of relevancy.
For the missions, real-time updates to flight paths were made based on fused sensor tracks proffered by virtual Advanced Framework for Simulation, Integration, and Modeling (AFSIM) models, and RL agent missions were dynamically selected by operators while the aircraft was in flight, demonstrating live, effective human-machine teaming for autonomy. This real-time operational data describing the performance of AI pilots will be put into GA-rapid ASI’s retention process for analysis and used to enhance future agent performance.
Michael Atwood, senior director of advanced programmes at GA-ASI, stated that the principles exhibited by these flights set the bar for operationally relevant mission system capabilities on CCA platforms. As the business continues to operationalise autonomy for CCAs, the combination of airborne high-performance computing, sensor fusion, human-machine teaming, and AI pilots making judgements at the speed of relevance demonstrates how rapidly GA-ASI’s capabilities are growing.
The researchers utilised a Collaborative Operations in Denied Environment (CODE) autonomous engine and the US government-standard OMS messaging protocol to allow communication between the RL agents and the LVC system. Utilising government standards like OMS will allow for the rapid incorporation of autonomy for CCAs.
Additionally, GA-ASI utilised an EMC2 from General Dynamics Mission Systems to operate the autonomy architecture. EMC2 is an open architecture Multi-Function Processor with a multi-level security infrastructure that is used to host the autonomy architecture, demonstrating the ability to bring high-performance computing resources to CCAs in order to execute rapidly adaptable mission sets based on the operational environment.
This is the latest in a series of ongoing autonomous flights conducted with internal research and development money to validate critical AI/ML principles for UAS.
