Enhancing Geomagnetic Navigation with PPO-LSTM: Robust Navigation Utilizing Observed Geomagnetic Field Data.

Area of Science:

• Robotics and Autonomous Systems
• Geophysics
• Artificial Intelligence

Background:

• Geospatial navigation in GPS-denied environments is challenging for autonomous vehicles.
• The Earth's geomagnetic field offers a potential alternative navigation resource.
• Incomplete sensor data necessitates advanced navigation strategies.

Purpose of the Study:

• To develop a robust navigation system for autonomous vehicles in GPS-denied environments.
• To leverage the Earth's geomagnetic field for reliable geospatial navigation.
• To address the challenges of partial observability and uncertainty in geomagnetic navigation.

Main Methods:

• Formulating the navigation problem as a partially observed Markov decision process (POMDP).
• Employing proximal policy optimization with long short-term memory (PPO-LSTM), a deep reinforcement learning framework.
• Validating the approach using real-world geomagnetic data from the International Geomagnetic Reference Field (IGRF) model under noisy conditions.

Main Results:

• PPO-LSTM demonstrated superior performance compared to baseline algorithms.
• The proposed method achieved smoother vehicle trajectories.
• Higher heading accuracy was observed, indicating improved navigation precision.

Conclusions:

• The PPO-LSTM framework effectively handles uncertainty and partial observability in geomagnetic navigation.
• This approach enables robust policies that adapt to complex geomagnetic gradients.
• It offers a viable and robust solution for geospatial navigation in challenging environments.