A Transformer Encoder Approach for Localization Reconstruction During GPS Outages from an IMU and GPS-Based Sensor

Area of Science:

• Robotics
• Geomatics Engineering
• Artificial Intelligence

Background:

• Accurate localization is essential for many applications, but Global Positioning System (GPS) signals are unreliable in weak signal or outage environments.
• Existing solutions primarily focus on real-time localization, leaving a gap for non-real-time applications requiring robust post-processing.
• High-velocity and acceleration scenarios, like tracking animals, present unique localization challenges not fully addressed by current methods.

Purpose of the Study:

• To develop and evaluate a novel post-processing localization method using a Transformer-based bidirectional encoder.
• To address the limitations of GPS during weak signal or outage periods for non-real-time applications.
• To improve localization accuracy and trajectory estimation in challenging dynamic environments.

Main Methods:

• A Transformer-based bidirectional encoder was employed to predict velocity during GPS signal degradation.
• Bidirectional velocity integration was utilized for position calculation, complemented by position interpolation for seamless transitions.
• The method was tested on a dataset involving high-velocity animal tracking.

Main Results:

• The proposed method achieved an average trajectory error below 3 meters in post-processing.
• Stable relative distance errors were maintained irrespective of Global Positioning System (GPS) outage duration.
• The approach demonstrated effectiveness in challenging datasets with velocities up to 10 times higher than pedestrians.

Conclusions:

• The Transformer-based bidirectional encoder offers a robust solution for post-processing localization during GPS outages.
• The method provides accurate and smooth trajectory estimation, particularly in high-dynamic scenarios.
• This approach enhances localization capabilities for applications where real-time constraints are not critical.