Ren Meilin, Wei Junyu, Qin Jiangyi, Guo Xiaojun, Wang Haowen, Li Shiqi
CATARC Automotive Test Center (Tianjin) Co., Ltd., Tianjin, 300300, China.
College of Intelligence Science and Technoloy, National University of Defense Technology, Changsha, 410000, China.
Sci Rep. 2025 Jul 1;15(1):21637. doi: 10.1038/s41598-025-05501-3.
This paper proposes a novel UWB/INS integration framework that utilizes attention-based Long Short-Term Memory (LSTM) neural networks to address challenges related to UWB signal degradation during non-line-of-sight (NLOS) propagation. The network is adopted to generate pseudo measurements to maintain Kalman filter measurement update during NLOS. LSTM networks are well-suited for modeling sequential data due to their ability to capture long-term dependencies, making them particularly effective in handling the temporal aspects of navigation data. By leveraging attention mechanisms, the proposed approach enhances temporal feature extraction and improves the accuracy of pseudo-UWB observations generation. Extensive experiments demonstrate that the attention-LSTM model significantly reduces positioning errors under both loosely and tightly coupled configurations in NLOS scenarios. This hybrid fusion of model-based and learning-based techniques ensures robust and precise UWB/INS localization.
本文提出了一种新颖的超宽带/惯性导航系统（UWB/INS）集成框架，该框架利用基于注意力机制的长短期记忆（LSTM）神经网络来应对非视距（NLOS）传播期间超宽带信号退化相关的挑战。该网络用于生成伪测量值，以在非视距情况下维持卡尔曼滤波器的测量更新。LSTM网络由于能够捕捉长期依赖关系，非常适合对序列数据进行建模，这使得它们在处理导航数据的时间特性方面特别有效。通过利用注意力机制，所提出的方法增强了时间特征提取，并提高了伪超宽带观测生成的准确性。大量实验表明，在NLOS场景中，注意力-LSTM模型在松耦合和紧耦合配置下均显著降低了定位误差。这种基于模型和基于学习的技术的混合融合确保了强大而精确的UWB/INS定位。
