accelerometer-guided separation

This study addresses the security risks associated with reconstructing CNC machine tool axis positions from accelerometer signals, a task where conventional double integration methods struggle to accurately recover trajectories under noisy conditions and complex operational scenarios. To overcome these limitations, this work proposes a novel sequence-to-sequence learning model based on Long Short-Term Memory (LSTM) networks that directly reconstructs high-precision axis and tool positions from broadband acceleration signals collected during industrial machining processes. The proposed approach significantly outperforms traditional methods, reducing reconstruction errors by 98% in low-complexity motions and by 85% in complex machining sequences, while preserving critical geometric features of the toolpath. These results demonstrate the feasibility of inferring sensitive positional information from readily available condition monitoring data, thereby revealing a previously underappreciated security vulnerability in industrial control systems.

This study addresses the challenge of motion artifacts corrupting wrist-based photoplethysmography (PPG) signals during daily activities, which degrades the accuracy of heart rate and respiratory rate estimation. The authors propose an accelerometer-guided, physics-constrained harmonic separation framework that models physiological signals through an analysis–synthesis process. By leveraging a physically interpretable harmonic generation mechanism, the method decouples cardiac and respiratory modulation components, while incorporating uncertainty-aware weighting to optimize signal reconstruction and enhance robustness to motion. Evaluated on the PPG-DaLiA dataset, the approach significantly outperforms existing methods, achieving highly accurate and interpretable joint estimation of heart rate and respiratory rate, and effectively separating physiological signals from motion-induced artifacts.

This paper addresses the problem of inferring an unknown number and positions of circular landmarks from imprecise binary sensory signals—generated solely by nearby landmarks—as an agent moves along a circle. The core challenge lies in modeling the signal as a piecewise-constant function on the circle, whose discontinuities correspond to landmark boundaries; however, the function values at discontinuities are unknown, and landmark radii, centers, and cardinality are all unknown. To tackle this, we introduce the notion of “robust decomposition”: a unique representation of the signal as a sum of restrictions to the circle of circular indicator functions. We characterize the solution space via the concept of “degrees of freedom.” Theoretically, we provide necessary and sufficient conditions for robust decomposability and a complete structural characterization. Algorithmically, we devise a deterministic procedure that enumerates all robust decompositions. Furthermore, we establish tight upper and lower bounds on the number of maximum-degree-of-freedom decompositions. Our results enable robust localization, obstacle avoidance, and motion planning in unknown environments without prior map knowledge.

Sparse inertial motion capture (MoCap) systems suffer from pose estimation drift under magnetic interference, limiting their practical deployment. To address this, we propose a “detect-then-correct” framework: first, real-time and precise magnetic interference detection via collaborative multi-IMU signal analysis; second, dynamic pose error correction within a sensor fusion pipeline by integrating a learned human motion prior model. The method requires no additional hardware and is plug-and-play compatible with mainstream sparse inertial MoCap systems. Experimental evaluation under representative magnetic interference scenarios demonstrates an average pose error reduction of 42.7%, significantly outperforming state-of-the-art approaches in both robustness and generalization. Our approach establishes a new paradigm for high-fidelity motion capture in complex electromagnetic environments.

This work addresses the challenge of real-time motion detection in streaming video—particularly under low signal-to-noise ratio and dynamic illumination conditions typical of simulated surveillance scenarios. We propose a lightweight, unsupervised detection framework based on Dynamic Mode Decomposition (DMD). Our method models video frame sequences as linear dynamical systems and establishes, for the first time, an interpretable mapping between foreground motion characteristics and the temporal evolution of DMD eigenvalues. By analyzing the eigenvalue spectrum to identify salient motion responses, and integrating sliding-window segmentation with ROC-driven adaptive thresholding, we achieve efficient online detection. Evaluated on a simulated surveillance dataset, our approach achieves an AUC exceeding 0.92, demonstrating high accuracy, low latency, and strong robustness. This work introduces a novel paradigm for unsupervised, real-time motion detection grounded in spectral system identification.

This work addresses the inherent limitation of conventional bearing-only methods, which rely on sensor lateral motion to ensure observability and thus face practical deployment constraints. To overcome this, the paper proposes a novel passive motion state estimation approach that fuses bearing measurements with received signal strength (RSS) information. By systematically leveraging RSS—a readily available yet long-overlooked modality—the method substantially enhances system observability and entirely eliminates the dependence on lateral sensor motion. Through rigorous observability analysis and nonlinear state estimation techniques, the proposed framework achieves accurate estimation of an unknown emitter’s motion state without requiring additional hardware. Experimental validation using real-world data confirms the efficacy and practicality of the approach.

This work addresses the vulnerability of state estimators to unmodeled disturbances—such as sensor aliasing or out-of-distribution noise—and their lack of a general self-assessment mechanism. To this end, the authors propose a sensor-agnostic introspective method that, for the first time, leverages frequency-domain spectral analysis for generic estimator health monitoring. By examining power spectral characteristics—such as signal power, bandwidth, and entropy—of recent velocity estimates, the approach operates without reliance on specific sensor models or assumptions about training data distributions. Experimental results demonstrate that this lightweight method is effective across diverse visual-inertial, LiDAR-inertial, and radar-inertial odometry systems, achieving 51%–58% fault recall and 60%–84% precision on real-world flight datasets, thereby validating the discriminative power of spectral features in detecting estimation failure.

Sensor measurements are often corrupted by outliers and non-Gaussian noise, leading conventional state estimators to produce biased and unreliable estimates. This work proposes an adaptive joint state and covariance estimation framework that uniquely integrates robust loss functions with covariance estimation. By combining norm-aware adaptive robust losses, iteratively reweighted least squares for state updates, and minimum weighted covariance determinant estimation within a block coordinate descent scheme, the method achieves self-tuning estimation without manual parameter tuning. Experimental results demonstrate that the approach accurately recovers inlier covariance in both Monte Carlo simulations and real-world ultra-wideband localization scenarios, achieving state estimation accuracy that matches or surpasses existing baseline methods.

This study addresses key challenges in bearing fault diagnosis—namely, the trade-off between global statistical features and local transient signals, insufficient physical interpretability of extracted features, and inefficient multi-source information fusion—by proposing a progressive, physics-guided multi-scale vibration signal processing framework. Integrating bearing kinematic theory with defect characteristic frequencies, the method constructs an 81-dimensional physically traceable feature space and introduces a fault-adaptive signal segmentation mechanism. Structured fault mechanism knowledge is implicitly encoded within a large language model architecture, enabling autonomous multi-scale fusion without external dependencies. Evaluated on four public datasets, the approach achieves 98.49% accuracy while reducing computational cost by 12.6× compared to baseline methods, with feature activations showing strong alignment with established fault mechanisms.

Traditional Kalman filtering is constrained by linear Gaussian assumptions, leading to suboptimal performance in nonlinear sensing scenarios such as Doppler radar and LiDAR, where mere parameter tuning cannot overcome inherent structural limitations. This work proposes the Kalman Evolve framework, which for the first time introduces algorithmic structure discovery into state estimation by jointly optimizing noise parameters and update structures. Leveraging large language models as structured priors over program space, the method employs program synthesis to generate interpretable, non-affine filtering algorithms that retain recursive form while adapting effectively to nonlinear dynamics. Experiments across diverse real-world and synthetic tracking tasks demonstrate up to a 12% reduction in root mean square error (RMSE), significantly outperforming strong existing baselines.