WRIGHT-PATTERSON AIR FORCE BASE, OHIO -- In a show of artificial intelligence in space, the Air Force Research Laboratory (AFRL) recently demonstrated autonomous control of a satellite bus using a neural network. The flight marks a pivotal shift from AI managing sensors and payloads to it directly commanding the critical function of spacecraft orientation.
During the mission, an AI agent was uplinked to a CubeSat orbiting hundreds of miles above Earth. Within a single orbit and without human intervention, the neural network autonomously controlled the spacecraft's orientation, precisely steering the vehicle to its mission objective and avoiding hazardous conditions.
"We have a responsibility to translate advances in AI to give our Airmen and Guardians trustworthy autonomous teammates to meet today's fast-moving challenges,” said Department of the Air Force Technology Executive Officer and AFRL Commander Brig. Gen. Douglas P. Wickert. “The AFRL vision is to win the future by executing our mission to discover, develop and deliver war-winning science and technology."
Brig. Gen. Douglas P. Wickert, Air Force Technology Executive Officer and AFRL Commander
AFRL's Autonomy Capability Team (ACT3) trained the neural network using reinforcement learning and subjected it to extensive pre-flight simulation and laboratory testing, bridging the gap between simulated environments and physical operations ahead of uploading it to the test vehicle.
“We used startup-like agility to take a reinforcement learning model from the lab directly to orbit,” said Dr. Steve “Cap” Rogers, AFRL senior scientist for AI enabled autonomy. “This flight is a perfect example of ACT3's core mission to operationalize AI at scale for the Air and Space Force.”
To ensure mission safety, the flight also tested a next-generation AI watchdog system and AFRL-designed guardrail runtime assurance systems. Watchdogs detect when AI systems approach unverified situations, enabling them to pivot seamlessly to backup systems. Guardrails monitor vehicle state and neural network outputs, intervening when predefined safety boundaries are approached. These systems ran in “shadow mode” and provided insight to how each would respond to the neural network in real time.
With the foundational capabilities of neural network control now proven in orbit, AFRL plans to expand the scope and complexity of autonomous operations for future flights.
About AFRL
The Air Force Research Laboratory is the Department of the Air Force’s primary scientific research and development center and one of six centers within Air Force Materiel Command. AFRL leads the discovery, development and delivery of technologies for air, space and the multidomain. With a workforce spanning seven mission areas at more than 40 locations worldwide, AFRL conducts research ranging from basic science to advanced technology development. For more information, visit: afresearchlab.com.
About ACT3
AFRL’s ACT3 is an AI Special Operations organization whose mission is to Operationalize AI at Scale for the Air Force and the Space Force. Commissioned by the AFRL Commander, ACT3 leverages an innovative ‘start-up’ business model as an alternative approach to the traditional AFRL Technical Directorate R&D model by combining the blue sky vision of an academic institution; the flexibility of an AI startup; and the discipline of a production development company. The goal of the ACT3 business model is to define the shortest path to successful transition of solutions using AFRL’s internal expertise and collaborations with the best academic and commercial AI researchers in the world. Successful implementation may mean new technology or new implementation of existing technology. To learn more about ACT3, visit: afrl.af.mil/ACT3/