Outcome Prediction and Explainability for Mission Operations of Autonomous Spacecraft

Alberto Candela, Tiago Vaquero, Bennett Huffman, Nihal Dhamani, Federico Rossi, Rebecca Castano

Keywords: spacecraft on-board autonomy, automated planning and execution, statistical machine learning, explainable artificial intelligence

Abstract: As planning and autonomy in general become increasingly deployed on board spacecraft, missions will face a paradigm shift in how ground operations teams command and interact with the spacecraft: moving from specifying timed sequences of commands to high-level goals that on-board autonomy will elaborate based on the spacecraft’s state and sensed environment. It will become increasingly difficult for operators to predict a mission's outcome as autonomous spacecraft venture into deeper space, react to unknown conditions, and face stronger communication constraints. However, data from simulated autonomous missions can be analyzed and leveraged, allowing operators to make informed decisions when selecting mission parameters and high-level goals. To this end, our paper presents a framework that gains insights from simulation data in order to help operators of autonomous spacecraft missions to predict, explain, and search for specific outcomes given a set of high-level goals. We show and discuss how our approach can help operators to better understand predictions, explore options and make informed decisions. To this end, we describe a case study that builds upon previous work on simulated autonomous spacecraft missions to the Neptune-Triton system where the spacecraft uses an automated planning and execution framework to make onboard decisions.