MLCA-A Machine Learning Framework for INS Coarse Alignment.

Area of Science:

• Navigation Systems
• Sensor Technology
• Machine Learning

Background:

• Inertial navigation systems (INS) require initial conditions for accurate dead-reckoning.
• Coarse alignment determines initial attitude using inertial sensors, crucial for navigation accuracy.
• Low-cost INS primarily use accelerometers for initial roll and pitch estimation.

Purpose of the Study:

• To develop a machine learning (ML) framework for stationary coarse alignment in INS.
• To improve the accuracy and efficiency of determining initial roll and pitch angles.
• To validate the proposed ML approach against conventional methods using simulations and real-world data.

Main Methods:

• A two-stage machine learning framework was developed for coarse alignment.
• Classical ML algorithms were employed to regress roll and pitch angles.
• The system was tested using both simulated data and field experiments with a smartphone.

Main Results:

• The proposed ML framework demonstrated effective coarse alignment of inertial sensors.
• The approach showed significant benefits compared to the standard analytical coarse alignment procedure.
• Validation through simulations and smartphone experiments confirmed the method's efficacy.

Conclusions:

• Machine learning offers a viable and advantageous alternative for coarse alignment in INS.
• The proposed framework enhances the accuracy of initial attitude determination, especially for low-cost systems.
• This research contributes to improving the performance of pure-inertial navigation by mitigating solution drift.