In various pedestrian motion scenarios, data-driven pedestrian dead reckoning (PDR) methods have demonstrated strong localization performance, significantly enhancing system adaptability and robustness. However, most existing deep learning models have high computational costs. This makes them difficult to deploy on energy-constrained mobile devices for long-term use. Current lightweight PDR networks adopt a single-input-single-output structure. Although effective in complex dynamics, this design becomes redundant in low-dynamic conditions such as static or slow walking. To address this challenge, we propose a lightweight network architecture named Motion-aware Early-exit Inertial Odometry Network (MEIO-Net). The model adopts a multi-exit design and incorporates a dedicated training scheme. Displacement covariance is used to assess motion difficulty. An early exit mechanism is then introduced to adaptively adjust the inference depth based on dynamic complexity. With a slight improvement in accuracy, MEIO-Net significantly reduces computational cost across multiple datasets. In the self-collected data set, it achieves a 48.9% reduction in FLOPs compared to state-of-the-art lightweight models. In typical simple pedestrian motion scenarios, the computational cost of the proposed model is reduced by 50.5%–80.5%. Ablation studies further confirm that MEIO-Net can flexibly terminate inference according to the intensity of the pedestrian motion. These results demonstrate its strong generalizability and practical potential for real-world deployment. Code and dataset are available at: https://github.com/SunXuehang/MEIO-Net.git.

The escalating complexity of network threats and the inherent class imbalance in traffic data present formidable challenges for modern Intrusion Detection Systems (IDS). While Graph Neural Networks (GNNs) excel in modeling topological structures and Temporal Convolutional Networks (TCNs) are proficient in capturing time-series dependencies, a framework that synergistically integrates both while explicitly addressing data imbalance remains an open challenge. This paper introduces a novel deep learning framework, named Gated Temporal Convolutional Network and Graph (GTCN-G), engineered to overcome these limitations. Our model uniquely fuses a Gated TCN (G-TCN) for extracting hierarchical temporal features from network flows with a Graph Convolutional Network (GCN) designed to learn from the underlying graph structure. The core innovation lies in the integration of a residual learning mechanism, implemented via a Graph Attention Network (GAT). This mechanism preserves original feature information through residual connections, which is critical for mitigating the class imbalance problem and enhancing detection sensitivity for rare malicious activities (minority classes). We conducted extensive experiments on two public benchmark datasets, UNSW-NB15 and ToN-IoT, to validate our approach. The empirical results demonstrate that the proposed GTCN-G model achieves state-ofthe-art performance, significantly outperforming existing baseline models in both binary and multi-class classification tasks.

The increasing integration of Industrial IoT (IIoT) exposes critical cyber-physical systems to sophisticated, multi-stage attacks that elude traditional defenses lacking contextual awareness. This paper introduces L2M-AID, a novel frame-work for Autonomous Industrial Defense using LLM-empowered, Multi-agent reinforcement learning. L2M-AID orchestrates a team of collaborative agents, each driven by a Large Language Model (LLM), to achieve adaptive and resilient security. The core innovation lies in the deep fusion of two AI paradigms: we leverage an LLM as a semantic bridge to translate vast, unstructured telemetry into a rich, contextual state representation, enabling agents to reason about adversary intent rather than merely matching patterns. This semantically-aware state empowers a Multi-Agent Reinforcement Learning (MARL) algorithm, MAPPO, to learn complex cooperative strategies. The MARL reward function is uniquely engineered to balance security objectives (threat neutralization) with operational imperatives, explicitly penalizing actions that disrupt physical process stability. To validate our approach, we conduct extensive experiments on the benchmark SWaT dataset and a novel synthetic dataset generated based on the MITRE ATT&CK for ICS framework. Results demonstrate that L2M-AID significantly outperforms traditional IDS, deep learning anomaly detectors, and single-agent RL baselines across key metrics, achieving a 97.2% detection rate while reducing false positives by over 80% and improving response times by a factor of four. Crucially, it demonstrates superior performance in maintaining physical process stability, presenting a robust new paradigm for securing critical national infrastructure.

Accurate positioning and navigation capabilities are essential for Internet of Things (IoT) devices. Event cameras, inspired by biological vision sensors, exhibit robust performance in high-dynamic and low-texture environments and are particularly suitable for IoT applications. However, it faces challenges with accuracy and scale in conventional slow-motion scenarios. Conversely, light detection and ranging (LiDAR) offers high precision in normal motion conditions but degrades significantly under high-dynamic motion. To integrate the advantages of both sensors, this article introduces the EVLINS algorithm, a multisource elastic fusion method based on an extended Kalman filter (EKF). This algorithm combines event-visual-inertial odometry (EVIO), LiDAR-inertial odometry (LIO), and an inertial measurement unit (IMU), utilizing a loosely coupled trajectory layer post-processing technique. This algorithm leverages the robustness of event cameras in highly dynamic environments and the precision of LiDAR in conventional settings, utilizing normalized uncertainty and nonholonomic constraint (NHC) strategies to address LIO’s degradation and EVIO’s accuracy issues. Thorough testing in various indoor and outdoor scenarios with real-world data demonstrates that EVLINS exhibits significantly improved accuracy and robustness compared to both LIO and EVIO algorithms. In large-scale, high-dynamic outdoor environments, EVLINS achieves a 3-D position accuracy of 0.68% over 1333.58 m, improving by 33.21% over LIO and 96.10% over EVIO, which diverged mid-way. In extreme indoor dynamic scenarios, EVLINS reduces maximum position error by 41.55% compared to LIO and improves overall position accuracy by 43.48%, and 22.96% compared to EVIO.

Multisensor integrated navigation, combining the global navigation satellite system (GNSS), inertial navigation system (INS), and light detection and ranging (LiDAR), is at the forefront of high-precision positioning technology. However, existing integration methods rely on raw point cloud data, which cannot be obtained in some projects due to geospatial data privacy concerns. For this problem, this work introduces the trajectory-layer-enhanced GNSS/INS/LiDAR integrated navigation system (TL-GILNS), which operates without raw point cloud data. This novel approach faces two primary challenges: 1) the difficulty in resolving the divergence of LiDAR positioning results due to cumulative errors and 2) the challenge of accurately assigning weights to LiDAR data in the integrated system. To overcome these obstacles, we propose a trajectory-layer LiDAR positioning enhancement method to reduce cumulative errors and a trajectory-layer LiDAR positioning accuracy quantification method to determine the weight of LiDAR. Finally, the performance of TL-GILNS is verified through experiments conducted in semi-open and large-scale complex scenes. In these scenes, TL-GILNS achieves horizontal accuracy better than 0.3 m, vertical accuracy better than 0.7 m, and yaw angle accuracy better than 1°. These results demonstrate the potential of TL-GILNS as a leading approach for high-precision navigation and positioning in complex environments.

LiDAR is widely used in Simultaneous Localization and Mapping (SLAM) and autonomous driving. The LiDAR odometry is of great importance in multi-sensor fusion. However, in some unstructured environments, the point cloud registration cannot constrain the poses of the LiDAR due to its sparse geometric features, which leads to the degeneracy of multi-sensor fusion accuracy. To address this problem, we propose a novel real-time approach to sense and compensate for the degeneracy of LiDAR. Firstly, this paper introduces the degeneracy factor with clear meaning, which can measure the degeneracy of LiDAR. Then, the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering method adaptively perceives the degeneracy with better environmental generalization. Finally, the degeneracy perception results are utilized to fuse LiDAR and IMU, thus effectively resisting degeneracy effects. Experiments on our dataset show the method’s high accuracy and robustness and validate our algorithm’s adaptability to different environments and LiDAR scanning modalities.

As a commonly used indoor Internet of Things (IoT) positioning method, fingerprinting is frequently carried out by fusing inertial and magnetic data. However, magnetic signals may exhibit high similarity in extensive indoor environments, leading to increased mismatching in magnetic fingerprinting positioning results. It negatively impacts the overall accuracy of positioning outcomes, leading to inaccuracies. To address this challenge, this article presents a multilevel error elimination approach that refines magnetic positioning outcomes from coarse to fine-grained adjustments. It marks the inaugural research on error detection and elimination specifically focused on magnetic field positioning. The method employs velocity information, sliding median filtering, and neighborhood filtering to rapidly, accurately, and effectively detect and eliminate the errors of magnetic positioning results. This methodology primarily employs velocity information to rapidly and comprehensively eliminate coarse errors. Subsequently, sliding median filtering addresses scenarios where velocity-based error correction is unreliable, effectively facilitating the intermediate removal of errors in magnetic field positioning outcomes. Ultimately, neighborhood filtering addresses unreliable situations in the above processes, enabling small-scale and detailed elimination of errors in magnetic field positioning results. Within a 100 m$\times $60 m indoor parking lot, the proposed approaches retained 51% to 80% of magnetic positioning results and enhanced the accuracy of magnetic positioning by over 80%.