A new preprint reports a method that builds lower and upper disturbance-reserve bounds, reaches exactness after finite expansions and delivers benchmark speedups over warm-started full solves.
A redesigned estimator separates measurement-dependent and grid-dependent work, cutting reported runtime and memory demands while retaining similar simulated error performance.
A preprint reports that dynamic sensor selection and time-frequency allocation outperformed equal resource sharing in a simulated cloud radar network when communication capacity was tight.
A mathematical framework for heterogeneous, constrained agents reached consensus in reported simulations and had the lowest listed cost in comparisons with 50 and 100 robots.
A computational study reports that clustering simulated grid conditions by their voltage responses reduced 8,760 hourly operating points to a 30-point set while reproducing steady-state voltage behavior with 98.3% accuracy. It also found lower reconstruction errors than injection-space clustering and heuristic sampling in the reported outage test.
A preprint proposes height- and elevation-aware analytical models for urban terrestrial-to-NTN links. In synthetic scenes and a stated simulation, the model produced higher coupling gains than the current 3GPP model, but the work does not establish real-world network performance.
A theoretical framework links operator inequalities with finite-dimensional observer design for changing Vlasov-Maxwell dynamics, but its numerical demonstration is small and solver-limited.
A new methods preprint finds that changing known input gains can expose hidden feedback links from outputs alone, while indirect paths and finite-sample inference limit broader structural recovery.
A neural network trained from power-flow objectives and constraints reported low violations and mostly small optimality gaps on five simulated feeders, with the largest reported gap at 1.5% on the 8500-node case.
A freight-corridor simulation reported higher throughput and lower congestion in its Cognitive scenario than in a manual Baseline, but also the highest total CO2 emissions proxy.
An arXiv methods preprint reports a median evaluation time of 2 microseconds for CIPS versus 121 microseconds for LA-ETC in an embedded microbenchmark. Its formal safety theorem is conditional on stated assumptions, while displayed car-following sweeps reported no CIPS fallbacks or violations.
A preprint reports that MVE led three comparison methods on historical trust measures in a simulation and was also tested under uncertainty and conflicting observations.
A preprint reports that TrustFormer produced lower trust-evaluation error and higher collaborator-selection accuracy than two baseline models in simulations of collaborative computing. The reported advantage was larger for strategic and gradually degrading behavior, while disorder in historical monitoring records increased error.
A four-case simulation of an energy-storage unit, EV fleet and thermostatically controlled load reported higher regulation scores and ancillary-service profits in a joint virtual power plant under fast-response requirements, but the result was limited to the modeled setup.
The model treats operating time as continuous and writes the result as a linear program. Its reported errors were measured against a two-minute reference model, but the paper's table and nearby prose assign one error value differently.
A modeling study proposes dynamic-droop checks for inverter controls and shows why a one-cycle power-response test can miss important differences between grid-forming and grid-following behavior.
A modeling study reports sufficient gain conditions for a dynamic-key LWE-encrypted controller to remain asymptotically stable and avoid overflow in one simulated linear control system. The result depends on stated assumptions and does not validate the design on physical equipment.
A preprint outlines three conditional control designs for hiding a selected network mode from compromised-node measurements. In one 11-node simulation, observer-based estimation converged where the output-feedback construction failed.
Movable antennas outperformed fixed-antenna PASS and conventional MIMO in the paper's simulated tests, while two specialized learning methods beat standard PPO across low-, medium- and high-security modes.
A 26 August 2026 arXiv preprint reports lateral drift in simulation and directional displacement in controlled indoor and outdoor flights during gyroscope-perturbation tests. The work uses a PX4-style stack and constrained sensor tests, but does not demonstrate fully detached, black-box, moving-target free flight.
The methods preprint compares a geometry-adaptive certified reduction with the full formulation in one closed-loop scenario and synthetic timing sweeps, reporting matching measures and conditional theorem-based guarantees.
In modeled agricultural grids, a GPU-based solver reported the lowest flight-time cost among three methods and kept reported runtime below 320 milliseconds on the tested dimensions. The study is a computational preprint, not evidence from a physical drone, field deployment or agricultural outcome.
A preprint reports a blockage-aware method that identified active RIS panels with more than 95% probability in simulated tests and achieved higher weighted sum rates than comparison methods under 30% per-link blockage.
A computational study describes a modular image-classification system that selects and compresses task-relevant visual information before sending it from a lightweight client to a larger server.
A preprint evaluates a multimodal system that forecasts future mmWave beams and selects virtual beam coverage, with a low-outage anchor result but sharply weaker performance on held-out scenarios.
A modeling study proposes a geometry-based boundary between near-field and far-field operation, then uses a Kronecker-SVD design to reduce beamforming cost while retaining near-full-SVD rates.
A hybrid method that pairs learned energy budgeting with convex optimization recorded the highest simulated computational throughput among the tested algorithms.
A model of space-based wireless power found quadratic received-power scaling under coherent satellite coordination, while non-coherent links showed linear scaling.
A conceptual preprint outlines three architectures for combining sensing observations from multiple UEs at a common base station, while leaving performance, scaling and deployment questions unresolved.
A small engineering evaluation of commodity Wi-Fi respiratory sensing reported 97% accuracy in normal breathing and 88% accuracy during detected pause episodes.
A 318 GHz preprint based on one factory hall found that strong-reflection clusters were fewer but denser, while random clusters were more numerous and more variable.
In mathematical and Monte Carlo simulations, a semi-blind channel estimator required fewer than half as many pilot symbols as a training-based method to reach the same target error. It also had lower plotted channel MSE than the tested EM and subspace-based methods, but the evidence is limited to modeled systems.
An engineering preprint reports similar simulated traction-control behavior from a Koopman-based controller, with much shorter solve times than nonlinear MPC.
Proc. 2024 28th Int. Conf. Syst. Theory Control Comput. (ICSTCC), 2024, pp. 26-314 min read
A computational preprint reports that UNION had low average cost gaps and high constraint satisfaction across seven AC-OPF systems, with selected outage tests and a five-day temporal evaluation.
A laboratory prototype combined digital communication and deformation sensing in one stretchable interconnect, with tests of transmission, strain response, repeated stretching and shape mapping.
A new methods preprint reports that intrinsic diffusion samplers respected manifold constraints in tests ranging from spherical distributions to robust rotation estimation, while leaving key questions about approximation error and broad performance unanswered.
Simulations report accurate DCT-based OFDM channel estimates and effective compensation with limited training, while nonlinear memory remains outside the model.
A methods preprint reports a live camera-aligned display of WiFi beamspace power, relative delay and polarization, with demonstrations separating direct, reflected and corner-reflected paths.
In a controlled synthetic test, attention-based ECG–PPG fusion had the lowest descriptive overall error, while signal-quality conditioning was most useful when PPG was completely missing. The comparisons used five training seeds, and none survived Holm correction.