| dc.description.abstract | As the number of objects in orbit grows, so does the risk of collisions. The sheer volume of collision warning messages far exceeds the capacity of human analysts, placing a significant burden on satellite operators and underscoring the need for autonomous, decentralized traffic management. Unlike centralized conjunction analysis, decentralized space traffic management distributes coordination across multiple independent nodes, allowing satellites to collaborate directly. This approach could enhance the resilience, speed, and international cooperation of space operations, helping to manage the space environment. For decentralized space traffic management to be viable, satellites must possess an accurate understanding of both the locations and intentions of other satellites. While satellites have precise knowledge of their own state, this accuracy diminishes when predicting the state of others. This gap is due to the limitations of onboard measurement systems and knowledge of each satellite’s structure, configuration, and maneuverability. Such differences motivate the exploration of information sharing between operators to improve coordination. Sharing information could benefit both individual operators and the broader space community by enabling more accurate trajectory predictions, facilitating formal maneuver negotiations, and enhancing overall orbital safety and efficiency. The main contribution of this thesis is to develop methods for autonomous satellite decision-making. By advancing the state of satellite autonomy, we can enhance high-level decision-making processes, enabling more adaptive and intelligent satellite coordination. This thesis begins by developing a multi-agent reinforcement learning environment to simulate satellite interactions in complex, high-dimensional settings. Then, we relax the assumption on synchronous communications and explore an alternate learning framework that relies on asynchronous communication between satellites. Our final contribution lies in a game-theoretic model of operator behavior in non-cooperative settings. Space is a competitive environment, and willingness to collaborate is mixed. As a result, we use game theory to obtain strategies to determine maneuvering and timing. | |
