A Machine Learning Approach for an Improved Inertial Navigation System Solution.

Area of Science:

• Navigation Systems
• Machine Learning
• Sensor Fusion

Background:

• Inertial Navigation Systems (INS) are crucial for continuous navigation but suffer from accumulating errors.
• Low-cost micro-electro-mechanical systems (MEMS) Inertial Measurement Units (IMUs) introduce significant errors like bias and noise, degrading INS performance.
• Existing methods for IMU error mitigation have limitations.

Purpose of the Study:

• To propose a novel machine-learning-based adaptive neuro-fuzzy inference system (ML-ANFIS) to enhance the performance of low-grade IMUs in INS applications.
• To develop an effective error model for low-grade IMUs by leveraging data from high-end IMUs.
• To validate the proposed algorithm's effectiveness in improving INS accuracy.

Main Methods:

• A two-phase approach was employed: training an ML-ANFIS model using 50% of low-grade IMU data alongside high-end IMU data to create an error model.
• The developed ML-ANFIS model was then tested on the remaining 50% of low-grade IMU measurements.
• Performance was evaluated using a real-world road trajectory.

Main Results:

• The proposed ML-ANFIS algorithm effectively mitigated errors from low-grade IMUs.
• Significant improvements were observed in the INS solution compared to traditional methods.
• A 70% enhancement in 2D positioning and a 92% enhancement in 2D velocity accuracy were achieved.

Conclusions:

• The ML-ANFIS algorithm demonstrates high effectiveness in correcting IMU errors and improving INS solutions.
• This approach offers a viable method for enhancing the usability of low-cost IMUs in navigation.
• The study highlights the potential of machine learning techniques in advancing navigation system accuracy.