Researchers have developed a new transformer-based architecture for managing resources in heterogeneous satellite clusters conducting Earth Observation missions, utilizing model-free reinforcement learning for real-time decision-making. This innovation promises improved adaptability and performance over traditional optimization methods, potentially transforming satellite operations in dynamic environments.