An Aerocapture Guidance and Estimation Framework for Improved Robustness to Uncertainty

Abstract

Aerocapture is an orbital insertion maneuver that converts a hyperbolic approach trajectory into a desired captured orbit using the aerodynamic forces generated during a single atmospheric pass. While it offers major benefits, such as reduced interplanetary cruise time and lower propellant mass reserves, it also introduces significant risk due to extreme sensitivity to atmospheric and delivery state uncertainties. This drives the need for robust guidance algorithms and accurate environmental estimation techniques. This thesis presents approaches to address both of these needs, developing solutions to improve aerocapture performance and robustness to uncertainty. The first contribution is the development of ABAMGuid+, a novel aerocapture guidance algorithm that leverages simultaneous control over bank angle and angle of attack. Inspired by optimal control theory, the algorithm uses a four-phase structure to mimic the optimal control laws while maintaining tractability for online use. Optimal control theory is utilized to identify the optimal control solutions, and numerical optimization is used to validate the analytic solutions prior to integration into a guidance algorithm. Extensive simulation results of a Uranus aerocapture scenario, including over 140,000 Monte Carlo trajectories, demonstrate significant improvements in capture success rates and propellant efficiency compared to existing methods. The second contribution addresses environmental uncertainty directly by developing a deep learning-based approach to estimate the atmospheric density profile during flight. A long short-term memory (LSTM) neural network-based architecture is trained to predict atmospheric density given sequences of flight data. The trained model is integrated into the guidance loop and a curriculum learning process is used to refine in-flight performance. Monte Carlo results show that the LSTM-augmented guidance system reduces propellant usage compared to traditional estimation methods. In summary, this thesis presents two approaches that improve aerocapture performance and robustness to uncertainty. We show that this added robustness can be achieved both by expanding algorithmic ability and by improving environmental estimation approaches.