Precision Mars Entry Navigation with Atmospheric Density Adaptation via Neural Networks

TL;DR

This work tackles the challenge of precise Martian entry navigation under atmospheric-density uncertainty by proposing an online density-adaptation approach using a neural network trained offline on a simple exponential density model. The network parameters are updated in real time through a maximum-likelihood framework that leverages measurement innovations and includes consider analysis to account for density uncertainty. Compared against covariance-matching and state-augmentation methods, the neural-network–augmented unscented Schmidt-Kalman filter (USKF-NN) delivers superior state-estimation accuracy and close agreement of the onboard density to Mars-GRAM profiles, achieving sub-percent density-tracking errors and consistent, unbiased navigation performance. This approach promises robust, computation-conscious online adaptation for EDL navigation in highly variable Martian atmospheres and can accommodate modular offline training with higher-fidelity models.

Abstract

Spacecraft entering Mars require precise navigation algorithms capable of accurately estimating the vehicle's position and velocity in dynamic and uncertain atmospheric environments. Discrepancies between the true Martian atmospheric density and the onboard density model can significantly impair the performance of spacecraft entry navigation filters. This work introduces a new approach to online filtering for Martian entry using a neural network to estimate atmospheric density and employing a consider analysis to account for the uncertainty in the estimate. The network is trained on an exponential atmospheric density model, and its parameters are dynamically adapted in real time to account for any mismatch between the true and estimated densities. The adaptation of the network is formulated as a maximum likelihood problem by leveraging the measurement innovations of the filter to identify optimal network parameters. Within the context of the maximum likelihood approach, incorporating a neural network enables the use of stochastic optimizers known for their efficiency in the machine learning domain. Performance comparisons are conducted against two online adaptive approaches, covariance matching and state augmentation and correction, in various realistic Martian entry navigation scenarios. The results show superior estimation accuracy compared to other approaches, and precise alignment of the estimated density with a broad selection of realistic Martian atmospheres sampled from perturbed Mars-GRAM data.

Figures (10)

• Figure 1: Position coordinates with respect to the MCMF frame $M$ (top) and velocity coordinates with respect to the Geographic frame $G$ (bottom).
• Figure 2: Altitude as a function of the density variation for the Mars-GRAM profiles generated. The gray lines show the different atmospheric profiles.
• Figure 3: Altitude (top) and density as obtained from the least squares exponential fit (bottom) as a function of time.
• Figure 4: Network architecture used to estimate the atmospheric density from the planet-centric radius.
• Figure 5: Probability histogram of the relative error between the predicted density from the trained neural network and the density in the validation set.