Attention based LSTM framework for robust UWB and INS integration in NLOS environments.

Area of Science:

• Navigation Systems Engineering
• Artificial Intelligence in Robotics
• Signal Processing

Background:

• Ultra-Wideband (UWB) and Inertial Navigation System (INS) integration is crucial for precise localization.
• Non-line-of-sight (NLOS) propagation significantly degrades UWB signal quality, impacting localization accuracy.
• Existing Kalman filter-based integration methods struggle with UWB signal degradation in NLOS environments.

Purpose of the Study:

• To propose a novel UWB/INS integration framework using attention-based LSTM neural networks.
• To address the challenge of UWB signal degradation in NLOS propagation.
• To enhance the robustness and accuracy of UWB/INS localization in challenging environments.

Main Methods:

• Development of an attention-based Long Short-Term Memory (LSTM) neural network.
• Utilizing the LSTM network to generate pseudo-measurements for Kalman filter updates during NLOS conditions.
• Implementing a hybrid fusion of model-based and learning-based techniques for UWB/INS integration.

Main Results:

• The attention-LSTM model effectively generates pseudo-UWB observations, maintaining Kalman filter updates.
• Significant reduction in positioning errors demonstrated in both loosely and tightly coupled UWB/INS configurations under NLOS scenarios.
• Enhanced temporal feature extraction and improved accuracy of pseudo-observations generation.

Conclusions:

• The proposed attention-LSTM framework offers a robust solution for UWB/INS localization in NLOS conditions.
• This hybrid approach effectively mitigates UWB signal degradation challenges.
• The method ensures precise and reliable localization through advanced deep learning integration.