This study presents an iterative numerical scheme for Helmholtz scattering with Neumann boundary conditions, modeling scattering from bounded convex bodies as a sum over propagation paths. The solution is expressed as the sum of the incident wavefield, specular reflection, and edge diffraction contributions. Recasting the Neumann series representation of the nth-order solution into tensor form provides a path interpretation and formally connects iterative path-tracing approaches to the known diffraction operator solution of the scattering problem. The study applies Nyström discretization to the nested diffraction integral, yielding reusable path diffraction coefficients. An iterative scheme is proposed that efficiently explores this path-tensor structure through successive expansions. The absolute value of the wavefield associated with each path serves as the ordering key in a max heap prioritization scheme. Numerical scattering experiments on the unit cube demonstrate rapid convergence. Relative L2-norm differences in the Dirichlet trace drop below 5%, 3.5%, and 3% after 10 000 iteration steps for wavenumbers k=2,4,6 m-1, respectively, when compared to results from direct boundary element formulations. For the k=2 m-1 case, which shows the largest trace error after 10 000 iteration steps, relative L2-norm errors of <2.5% and relative L∞-norm errors of <4% in the domain are observed.
Artificial intelligence (AI) powered tools are increasingly used in everyday activities and applications. AI-powered search engines effortlessly provide answers to numerous questions—from online shopping recommendations to medical diagnoses. This feedback seems ready for immediate application, which evokes an allure to use AI-powered tools to streamline a researcher’s laborious work. Can generative AI-powered search engines, leveraging advanced algorithms on vast datasets, be an instrument for accelerating in-spot data retrieval in the material sciences? As an example of AI’s applicability in material modelling, the efficiency of generative AI search engines in calibrating a material modelling approach is hereby investigated. As an example, a well-known and popular constitutive model, i.e., the Johnson–Cook plasticity model, has been chosen to estimate the behaviour of the aluminium alloy AA7020-T651 under a dynamic compression loading. The Johnson–Cook model describes the plastic behaviour of ductile materials under severe loading conditions, including large strains, dynamic strain rates, and elevated temperatures. Although the Johnson–Cook model is typically regarded as a relatively simple phenomenological model that can be fitted with a limited number of experimental tests (especially when compared to physically based constitutive formulations), deriving and validating the Johnson–Cook model still requires thorough experimental investigation and a comprehensive literature review. The current investigation is based on the outcomes generated by the three most popular generative AI platforms with freely accessible versions, which are hereby discussed, i.e., ChatGPT (by OpenAI), Gemini (by Google LLC), and Copilot (by Microsoft Corporation). Although generative AI-powered search engines can rapidly analyse literature, extract relevant data, and predict material responses under specified conditions, their output may result in incorrect or biased data due to searches within public sources, open-access research, and not necessarily professional scientific databases. Therefore, to confirm the ability of generative AI tools to deliver results aligned with the most recent evidence-based findings, a current study verifies how specific, efficient, and accurate the AI-generated response can be in application to a task that is objectively evaluable—fitting the Johnson–Cook plasticity model.
Object tracking from Unmanned Aerial Vehicles (UAVs) is challenged by platform dynamics, camera motion, and limited onboard resources. Existing visual trackers either lack robustness in complex scenarios or are too computationally demanding for real-time embedded use. We propose an Modular Asynchronous Tracking Architecture (MATA) that combines a transformer-based tracker with an Extended Kalman Filter, integrating ego-motion compensation from sparse optical flow and an object trajectory model. We further introduce a hardware-independent, embedded oriented evaluation protocol and a new metric called Normalized time to Failure (NT2F) to quantify how long a tracker can sustain a tracking sequence without external help. Experiments on UAV benchmarks, including an augmented UAV123 dataset with synthetic occlusions, show consistent improvements in Success and NT2F metrics across multiple tracking processing frequency. A ROS 2 implementation on a Nvidia Jetson AGX Orin confirms that the evaluation protocol more closely matches real-time performance on embedded systems.
Multi-source localization from time difference of arrival measurements in sensor networks is challenging in the presence of clutter, timing uncertainty, missed detections, and an unknown number of sources. We propose a combinatorial framework that jointly addresses these challenges by evaluating feasible subsets of detections, referred to as coalitions. Each coalition is assigned an associated cost defined as the $\ell_{\infty}$-norm of the time difference of arrival residuals. We retain coalitions whose residual lies below a user-specified pseudorange threshold, providing a physically interpretable control of coalition feasibility. Coalition selection is performed using a family of set packing algorithms, including greedy and binary integer linear programming formulations with cost-based and lexicographic objectives. The lexicographic variants maximize sensor consensus by prioritizing, from largest to smallest cardinality, the number of selected disjoint detection coalitions, and break ties by minimizing the sum of coalition costs. This provides a unified, geometrically consistent framework for joint data association based on maximized sensor consensus, clutter suppression, source counting, and localization. The methods are benchmarked using Monte Carlo simulations under varying levels of clutter, time jitter, and missed detections, as well as using an acoustic dataset recorded in an urban environment.
