Charles J. Wetterer, KBR; Jason Baldwin, Complex Futures LLC; Paul Billings, KBR; Keric Hill, KBR; Chris Craft, KBR; Juan Gutierrez, KBR; John Gaebler, KBR; Micah Dilley, KBR; Jill Bruer, AFRL
https://doi.org/10.64861/LBFK6030
Keywords: Orbit Determination, Cislunar, Estimation, Process Noise
Abstract:
When using a sequential estimation filter for orbit determination and maneuver detection, a process noise in the dynamics model is required to account for unmodeled or mismodeled effects. This is particularly true for objects in cislunar space where the dynamics model not only needs to include the point-mass gravitational influence of the Earth and Moon, but often higher-order aspheric gravitational effects, influences of other bodies (e.g., the Sun), and solar radiation pressure (SRP). Insufficient process noise will lead to filter overconfidence, unresponsiveness, and divergence, while excessive process noise will reduce the accuracy and precision of the state estimate and reduce sensitivity to maneuver detection. Regardless of the underlying process noise model being used, whether discretely applied or continuously applied, the required and sufficient magnitude of process noise is dependent on perhaps poorly known characteristics of the object itself (e.g., area-to-mass ratio, shape, attitude) and the object’s specific orbit and is essentially incalculable a priori. In this paper we develop an autonomous multiple model (AMM) approach for process noise tuning where multiple hypotheses using different process noise magnitudes are used in the estimation filter for the purpose of determining the best magnitude as part of the estimation process. This is done by tracking and comparing the weight gain for each AMM hypothesis as calculated in the measurement update step and establishing thresholds regarding when the nominal process noise power is switched, when the process noise powers are reset, and how these functions are prioritized as compared to the interacting multiple model (IMM) hypotheses used for maneuver detection. These AMM hypotheses, by construction, do not interact with each other or the IMM hypotheses, but they can inform the nominal IMM hypothesis with updated process noise powers as required. A variety of maneuvering and non-maneuvering objects in a wide range of orbits (e.g., geosynchronous Earth orbit (GEO), high-eccentricity Earth orbit (HEO), cislunar, Sun-Earth L2) are simulated to demonstrate how the AMM (for process noise tuning) and the IMM (for maneuver detection) react and affect filter results in situations with underestimated, overestimated, and nominal initial process noise powers.
Date of Conference: September 15-18, 2026
Track: Astrodynamics