The US has demonstrated a level of artificial intelligence in orbit after a neural network autonomously controlled the orientation of a satellite without human intervention.<\/p>\n
The flight, conducted by the Air Force Research Laboratory (AFRL), moved AI beyond managing sensors and payloads and placed it in charge of a core spacecraft function.<\/p>\n
During the mission, an AI agent was uploaded to a satellite<\/a> orbiting hundreds of miles above Earth and directed the vehicle toward its objective within a single orbit.<\/p>\n The system also avoided hazardous conditions while controlling the spacecraft\u2019s orientation, showing how autonomous software could reduce dependence on ground commands during military and scientific missions.<\/p>\n From payload management to spacecraft control<\/p>\n The demonstration represents a shift in how AI could be used in space operations. <\/p>\n Instead of supporting only mission equipment, the neural network directly commanded the satellite<\/a> bus, which determines how a spacecraft points and maneuvers in orbit.<\/p>\n \u201cWe have a responsibility to translate advances in AI to give our Airmen and Guardians trustworthy autonomous teammates to meet today\u2019s fast-moving challenges,\u201d said Department of the Air Force Technology Executive Officer and AFRL Commander<\/a> Brig. Gen. Douglas P. Wickert.<\/p>\n \u201cThe AFRL vision is to win the future by executing our mission to discover, develop and deliver war-winning science and technology,\u201d he added. <\/p>\n AFRL\u2019s Autonomy Capability Team, known as ACT3, developed the neural network using reinforcement learning. <\/p>\n The team trained the model through repeated simulations before exposing it to laboratory tests designed to narrow the gap between virtual performance and real spacecraft operations.<\/p>\n Reinforcement learning model reaches orbit<\/p>\n The project followed a rapid development approach intended to move AI systems from research environments into operational testing quickly.<\/p>\n \u201cWe used startup-like agility to take a reinforcement learning model from the lab directly to orbit,\u201d said Dr. Steve \u201cCap\u201d Rogers, AFRL senior scientist for AI enabled autonomy. \u201cThis flight is a perfect example of ACT3\u2019s core mission to operationalize AI at scale for the Air and Space Force.\u201d<\/p>\n Reinforcement learning allows an AI model to improve its decisions by receiving feedback from simulated actions and outcomes. For this mission, the approach enabled the agent to learn how to orient the spacecraft while remaining within operational limits.<\/p>\n The successful flight offers evidence that neural network-based control can operate on physical spacecraft after simulation and ground validation.<\/p>\n Safety systems monitor autonomous decisions<\/p>\n