Operations for Autonomous Spacecraft: Downlink Analysis of Onboard Decisions and Execution Anomalies

Future space exploration missions will heavily rely on autonomous planning and execution (APE) technology to improve spacecraft reliability and reduce operational costs. However, this will require a complete revamp of ground operations, i.e., from current practice of specifying pre-planned sequences to specifying high-level goals that will later be elaborated by the onboard APE based on spacecraft's state and perceived environment. Particularly, determining the mission outcomes during downlink is a challenging task. In this paper, we reconstruct what the spacecraft has executed onboard (i.e., as-executed) using downlinked channelized data, EVRs, and, critically, spacecraft models; We also quantitatively compare as-executed from the "actual" run with ground-based prediction simulations. To do this quantitative comparison, we design an N-dimensional dynamic time warping (DTW) technique based on which we formulate two similarity scores: (a) one related to executed tasks whose cost function is based on interval-based generalized intersection over union; and (b) other related to spacecraft states whose cost function is based on normalized Manhattan distance. Through a simulated case study of multiple flybys in NeptuneTriton system, we demonstrate that our technique successfully quantifies the similarity between the as-executed actual and predicts, and assess its "in-family" versus "out-of-family" behavior. To lower the associated false positives/negatives, we also design a multi-objective assessment metric that is a weighted summation of the task- and timeline-related similarity scores.