Benjamin L. Reifler, The University of Texas at Austin; Brandon A. Jones, The University of Texas at Austin
Keywords: multi-fidelity, cislunar, multi-target
Abstract:
Increasing human activity in cislunar space presents new challenges for space situational awareness (SSA). In particular, chaotic dynamics make orbit uncertainty propagation difficult, requiring significant computation time for sufficient accuracy and often resulting in rapid growth in uncertainty. The Moon also causes difficulties in detecting objects in its vicinity, which can result in sparse observations. As the population of objects in cislunar space grows, so will the need for efficient multi-target tracking algorithms that can mitigate these issues. The computational cost of a multi-target filter’s prediction step is generally proportional to the cost of predicting a single object’s probability density function (PDF) times the number of objects being tracked. Therefore, total filter runtime is increasingly dependent on the efficiency of orbit uncertainty propagation as the number of objects increases.
In order to reduce the computational cost of track prediction while maintaining accuracy, this paper presents a method for adaptive multi-fidelity orbit uncertainty propagation. The proposed method is applied to the ensemble Gaussian mixture filter (EnGMF), a particle-based single-target filter that has been shown to enable accurate space object tracking with sparse observations. We use the EnGMF with adaptive multi-fidelity propagation in a multi-target filter and assess the effect of our adaptive propagation method on tracking accuracy and computational cost in simulated cislunar SSA scenarios.
Multi-fidelity orbit uncertainty propagation balances the accuracy of high-fidelity propagation and the speed of low-fidelity propagation to increase the computational efficiency of particle-based filters. For example, bi-fidelity propagation works by propagating all particles with a relatively cheap low-fidelity model, identifying a small set of important particles, propagating these particles again with the more costly high-fidelity model, and correcting the low-fidelity particles to better match the distribution of the high-fidelity particles via stochastic collocation. The selection of the dynamics models determines the overall accuracy and computational cost of multi-fidelity propagation. The accuracy of multi-fidelity propagation can also be increased by using a time history of states for each particle to construct the multi-fidelity surrogate.
The varied orbital dynamics and regimes over cislunar and near-lunar space introduce new challenges for multi-fidelity uncertainty propagation not present in the near-Earth case. Depending on an object’s location in cislunar space, it is possible to select the optimal pair of dynamics models from a set of models to achieve a desired level of accuracy while minimizing computational cost. For example, if the object is close to the surface of the Moon, a detailed non-spherical gravity field model may be necessary, but if the object is farther from the surface, a point-mass model may be sufficient. If the object is far from both the Earth and Moon, the optimal model may be the circular restricted three-body problem (CRTBP) in a rotating reference frame or a model that includes point-mass gravity of the Earth and Moon.
In this paper, we apply our adaptive multi-fidelity propagation method to simulated SSA scenarios in which objects pass through a variety of orbital regimes with different dynamics. We use these scenarios to quantify the effect on multi-target filter runtime and tracking accuracy when using this method, with the goal of improving our ability to efficiently track the growing cislunar population.
Date of Conference: September 17-20, 2024
Track: Cislunar SDA