24 February 2025 in IEEE Transactions on Instrumentation and Measurement
Various animals exhibit accurate navigation using environmental cues. The Earth’s magnetic field has been proven to be a reliable information source for long-distance fauna migration. Inspired by animal navigation, this work proposes a bionic and data-driven approach for long-distance underwater navigation. The proposed approach uses measured geomagnetic data for navigation and requires no GPS systems or geographical maps. Particularly, we construct and train a Temporal Attention-based Long Short-Term Memory (TA-LSTM) network to predict the heading angle during navigation. To mitigate the impact of geomagnetic anomalies, we develop a mechanism to detect and quantify the anomalies based on Maximum Likelihood Estimation (MLE). We integrate the developed mechanism with the TA-LSTM and calibrate the predicted heading angles to gain resistance against geomagnetic anomalies. Using the retrieved data from the WMM model, we conduct numerical simulations with diversified navigation conditions to test our approach. The simulation results demonstrate the resilience of our navigation approach against geomagnetic anomalies, along with the precision and stability of underwater navigation in single and multiple destination missions.
17 October 2024 in IEEE Sensors Journal
Due to the absence of Global Navigation Satellite Systems (GNSS) in the underwater environment, navigation utilizing the Geophysical Magnetic Fields of the Earth (GMFE) emerges as a promising field. Geomagnetic gradient-based navigation systems rely on pre-mapping designated areas to minimize the effects of magnetic anomalies on navigation performance. However, obtaining magnetic maps beforehand is challenging, and relying solely on historical maps is insufficient for describing magnetic anomalies. Significant changes in magnetic field gradients caused by unidentified anomalies can render conventional geomagnetic navigation methods ineffective. In this article, we present a novel method called the Geomagnetic Gradient Multi-Attention Neural Network (GGMA). Our proposed encoder and decoder attention mechanism is capable of predicting the position and geomagnetic trends based on various geomagnetic gradient elements extracted from the trajectory data. We additionally utilize maximum likelihood estimation to ascertain the probability of the current coordinates being located within the anomalous region. By integrating the expected heading with the calculated heading, the effect of magnetic field anomalies on geomagnetic navigation is minimized. The GGMA method is effective in solving the navigation failure problem of magnetic field gradient-based navigation methods in the presence of magnetic field anomalies during long-distance simulation navigation experiments. It has higher navigation efficiency and accuracy compared to other algorithms in regions without anomalies.
13 August 2025 in IEEE Transactions on Instrumentation and Measurement
The advancement in autonomous vehicles has empowered navigation and exploration in unknown environments. Geomagnetic navigation for autonomous vehicles has drawn increasing attention with its independence from the global positioning system (GPS) or inertial navigation devices. While geomagnetic navigation approaches have been extensively investigated, the generalizability of learned geomagnetic navigation strategies remains unexplored. The performance of a learned strategy can degrade outside of its source domain where the strategy is learned, due to a lack of knowledge about the geomagnetic characteristics in newly entered areas. This article explores the generalization of learned geomagnetic navigation strategies via deep reinforcement learning (DRL). Particularly, we use DRL agents to learn multiple teacher models from distributed areas that represent dispersed navigation strategies and amalgamate the teacher models for generalizability across navigation areas. We design a reward shaping mechanism in training teacher models where we integrate both potential-based and intrinsic-motivated rewards. The designed reward shaping can enhance the exploration efficiency of the DRL agent and improve the representation of the teacher models. Upon the gained teacher models, we use multiteacher policy distillation to merge the policies learned by individual teachers, leading to a navigation strategy with generalizability across navigation areas. We conduct numerical simulations, and the results demonstrate an effective transfer of the learned DRL model from a source domain to new navigation areas. Compared with existing evolutionary-based geomagnetic navigation methods, our approach provides superior performance in terms of navigation length, duration, heading deviation, and success rate in cross-domain navigation.
11 August 2026 in Measurement
13 November 2026 in Measurement Science and Technology
To address the inherent challenges of localization in regions with weak or ambiguous magnetic gradients, this paper proposes a lightweight residual multilayer perceptron (MLRP) method for geomagnetic matching navigation. By establishing an accurate nonlinear mapping between geomagnetic vectors and spatial coordinates, the proposed method prevents gradient degradation and the consequent decline in accuracy and stability observed in existing geomagnetic matching methods. Furthermore, the MLRP method employs an efficient network architecture, achieving a balance between efficiency and accuracy on resource-constrained mobile platforms, making it suitable for long-distance navigation missions. Simulation results demonstrate that, under varying noise conditions, MLRP achieves the lowest mean route-level error among the evaluated methods. In one measured route in Xi’an, MLRP reduced by 13.94% relative to iterative closest contour point and achieved a batch-one latency of 1.7706 ms on an NVIDIA Jetson TX2, confirming the practical feasibility of achieving high-precision autonomous navigation in global navigation satellite system-denied environments using the MLRP method.