AI Tools for Design and Operation of Distributed Spacecraft Missions

With the recent advances in satellite miniaturization, communication and information technologies, there has been a paradigm shift in space exploration missions over the last few decades. This paradigm shift involves the transition from monolithic architectures formed by just one big satellite to a concept of a sensor web for space exploration consisting of heterogeneous sensors hosted on a variety of platforms including space, air and ground assets. These multiple entities share information in real time and make coordinated autonomous decisions to maximize system performance and/or scientific value. In this context, this thesis uses AI and machine learning techniques to overcome two big challenges found in the design and operation of Distributed Spacecraft Missions (DSMs): (1) The combinatorial explosion of feasible Earth observing constellations when not constraining the satellite orbits to symmetrical configurations, such as the Walker pattern. (2) The constant monitoring and ground operations required for node buffer management in Delay Tolerant Networks (DTN), which are governed by a set of standardized internet-like communications protocols robust to long delay and constant disruptions, and used in the communication between nodes in DSMs. The first challenge is approached by creating novel evolutionary formulations to explore large tradespaces of non-Walker hybrid satellite constellations with diversity of orbital parameters. Finally, the second challenge is addressed with the use of deep reinforcement learning, to automate the on-board decision making process in certain aspects of memory buffer management in DTN nodes, with the ultimate goal of optimizing network performance and reducing operational costs