Static Attitude Determination Using Convolutional Neural Networks.

Area of Science:

• Aerospace Engineering
• Computer Science
• Robotics

Background:

• Spacecraft navigation relies heavily on accurate estimation of orientation between reference frames.
• Traditional algebraic methods for attitude determination face challenges with measurement uncertainty.
• Data-driven approaches, particularly neural networks, show promise for handling stochastic data in navigation.

Purpose of the Study:

• To propose a novel approach for static attitude determination using convolutional neural networks (CNNs).
• To address and mitigate the impact of measurement uncertainty in attitude estimation.
• To compare the performance of the proposed CNN model against established algorithms.

Main Methods:

• Development and training of PointNet-based convolutional neural network models.
• Utilizing attitude profile matrices, derived from observation vectors, as input data.
• Evaluating model performance under various measurement noise conditions.

Main Results:

• The proposed CNN model demonstrated superior performance compared to traditional algorithms like SVD, q-method, QUEST, and ESOQ2.
• The CNN approach exhibited reduced sensitivity to higher levels of measurement noise.
• Model selection was guided by performance under tested uncertainty levels.

Conclusions:

• Convolutional neural networks offer a robust and noise-tolerant solution for static attitude determination in spacecraft.
• The data-driven method provides a viable alternative to algebraic approaches, especially in scenarios with significant measurement uncertainty.
• This work advances the application of AI in critical aerospace navigation systems.