Context Aware Transformer Network and In-situ IMU Calibration For Accurate Positioning

Common inertial localization approaches based on the inertial navigation system algorithm encounter challenges in accurately tracking complex movements, such as those occurring in smartphones or wearables, and tend to drift over a short period of time. This work aims to overcome these challenges through a state-of-the-art supervised learning approach. We use a novel transformer network to solve both the calibration and localization tasks simultaneously. The integration of physics-based features into the transformer model further improves the accuracy and reliability of the localization by providing context awareness for the sensor readings. In addition, the use of learning approaches as an in-situ calibration method would improve the accuracy of the estimation. By implementing Transformer-based attention mechanisms, we combine physics-based and learning-based methods, leading to an advanced solution that enables precise and reliable localization in complex device placements. Additionally, comprehensive numerical and analytical evaluations with realistic quality assessments using multiple metrics are presented.