Multi-unmanned aerial vehicle (UAV) systems provide scalable service platforms for large-scale environmental tasks, such as grassland ecosystem restoration. However, coordinating fleet operations requires solving the restoration area maximization problem (RAMP). This non-linear combinatorial optimization challenge is complicated by payload-dependent energy dynamics and heterogeneous ecological degradation. We propose a novel knowledge-guided collaborative bilevel formerpointer reinforcement learning framework (KC-BFPRL) to address this complexity. Using a hierarchical paradigm, KC-BFPRL decomposes RAMP into global task allocation and local restoration planning, with the latter further divided into upper-level trajectory planning and lower-level restoration area allocation. Our specialized architecture pairs featuring a Transformer-based encoder that fuses static environmental features with dynamic UAV states, and a Pointer Network decoder trained via a robust actor-critic framework. By embedding ecological priority rules and heuristic logic, KC-BFPRL achieves a structured warm-start, solving the RL cold-start problem while ensuring strict constraint satisfaction. Extensive experiments demonstrate that KC-BFPRL consistently outperforms state-of-the-art baselines, achieving superior objective values and efficiency. It maintains a 0.00% optimality gap in the most complex scenarios U8-R160 and operates nearly three times faster than MAPDP, validating its robustness, scalability, and real-time applicability for large-scale automated ecological restoration.

This paper proposes a novel robust tracking model predictive control (MPC) method within the Koopman framework for unknown nonlinear constrained systems. The key idea is to continuously update the Koopman surrogate model using new input-state measurements, while ensuring recursive feasibility of the MPC scheme under modeling errors through constraint tightening. The proposed method assumes that the modeling error is Lipschitz-continuous and estimates the corresponding Lipschitz constant from the training error in the dataset. Based on this estimate, a bound on the prediction error over the horizon is derived, leading to a convex robust optimal control formulation. Compared to existing Koopman-based approaches, the proposed method demonstrates superior modeling performance and improved data efficiency. Despite the need for online identification of the surrogate model and the construction of tightened constraint sets, the method remains real-time implementable. Furthermore, the input-to-state stability (ISS) of the closed-loop system is rigorously established. The effectiveness of the proposed method is demonstrated through numerical examples. (c) 2026 Published by Elsevier Ltd.

Unmanned aerial vehicles (UAVs) are increasingly used in smart city communications for air-ground communications due to their flexibility, low cost, and independence from ground conditions, enabling high data rates for future networks. This paper explores UAV-to-vehicle (U2V) mmWave integrated sensing and communication (ISAC), where vehicles are represented as rigid shapes in a 3D radar point cloud. Considering maximizing channel capacity with multi-user interference and radar performance, two adaptive optimization problems are proposed, incorporating vehicle-to-vehicle (V2V) communication for interference mitigation. The radar point cloud-driven reinforcement learning (PointRL) algorithm is designed to solve these problems. It includes a point cloud-based deep neural network (PDNN) for extracting action spaces from 3D radar data and a decision network that reduces network complexity through segmentation and connection. A linear weighted sliding window reward mechanism is also designed to enhance decision-making in dynamic environments. Simulation results show that the proposed PointRL outperforms benchmark methods.

This paper presents a safe reinforcement learning (RL) framework for multi-agent systems based on data-driven distributed robust model predictive control (D3RMPC). Our approach leverages a parameterized D3RMPC as a function approximator, optimizing closed-loop performance through online updates to determine optimal policies for each agent. Unlike traditional MPC and existing MPC-based RL methods, D3RMPC relies solely on past system data and an implicit model based on behavioral system theory, eliminating the need for an explicit state-space model. By combining D3RMPC with the RL approach, we achieve explainable safety guarantees while overcoming the challenges of model mismatch. A formal theoretical framework ensures the preservation of safety, stability, and feasibility throughout the system's learning process and closed-loop operation. The efficacy of the proposed method is demonstrated through two numerical simulations.

Orthogonal time frequency space (OTFS) modulation offers strong resilience to Doppler effects but suffers from high system latency, limiting its use in low-latency communications. This paper proposes a channel estimation algorithm for low-latency OTFS systems with large delay spreads and fractional Doppler effects. In the delay-time (DT) domain, impulse pilots are placed at equidistant intervals along the first row of the DT grid to eliminate interference from aliased delays, and a threshold detection method estimates the delays. Doppler shifts and path gains are then estimated using the discrete Fourier transform (DFT). Simulation results show that the proposed algorithm achieves near-optimal bit error rate (BER) under sparse delay spreads.

Calibrating Agent-Based Models (ABMs) is an important optimization problem for simulating the complex social systems, where the goal is to identify the optimal parameter of a given ABM by minimizing the discrepancy between the simulated data and the real-world observations. Unfortunately, it suffers from the extensive computational costs of iterative evaluations, which involves the expensive simulation with the candidate parameter. While Surrogate-Assisted Evolutionary Algorithms (SAEAs) have been widely adopted to alleviate the computational burden, existing methods face two key limitations: 1) surrogating the original evaluation function is hard due the nonlinear yet multi-modal nature of the ABMs, and 2) the commonly used surrogates cannot share the optimization experience among multiple calibration tasks, making the batched calibration less effective. To address these issues, this work proposes Automatic posterior transformation with Negatively Correlated Search and Adaptive Trust-Region (ANTR). ANTR first replaces the traditional surrogates with a pretrainable neural density estimator that directly models the posterior distribution of the parameters given observed data, thereby aligning the optimization objective with parameter-space accuracy. Furthermore, we incorporate a diversity-preserving search strategy to prevent premature convergence and an adaptive trust-region method to efficiently allocate computational resources. We take two representative ABM-based financial market simulators as the test bench as due to the high non-linearity. Experiments demonstrate that the proposed ANTR significantly outperforms conventional metaheuristics and state-of-the-art SAEAs in both calibration accuracy and computational efficiency, particularly in batch calibration scenarios across multiple market conditions.

Highlights What are the main findings? A unified graph-theoretic framework that integrates the environment, evasive targets, and UAVs, providing a consistent framework for cooperative search. Develops a Directional Gradient-based Prediction (DG-Prediction) model that captures target evasion intent by guiding probability diffusion along predicted escape corridors, moving beyond isotropic assumptions. What are the implications of the main findings? Enables effective search against intelligent evasive targets by providing a structured, intention-aware model that significantly improves target capture probability in adversarial scenarios. Delivers a practical, computationally efficient system that bridges the gap between discrete decision-making and continuous, kinematically feasible trajectory generation for fixed-wing UAV swarms.Highlights What are the main findings? A unified graph-theoretic framework that integrates the environment, evasive targets, and UAVs, providing a consistent framework for cooperative search. Develops a Directional Gradient-based Prediction (DG-Prediction) model that captures target evasion intent by guiding probability diffusion along predicted escape corridors, moving beyond isotropic assumptions. What are the implications of the main findings? Enables effective search against intelligent evasive targets by providing a structured, intention-aware model that significantly improves target capture probability in adversarial scenarios. Delivers a practical, computationally efficient system that bridges the gap between discrete decision-making and continuous, kinematically feasible trajectory generation for fixed-wing UAV swarms.Abstract Rapid search for evasive targets using multiple Unmanned Aerial Vehicles (UAVs) presents significant challenges, as it requires real-time target-motion prediction, multi-agent coordination, and adherence to kinematic constraints. Existing cooperative search methods often assume non-adversarial target behavior or model target motion independently of UAV actions, which reduces their effectiveness against targets that actively evade based on UAV positions. To address these limitations, this study introduces the Cooperative Rapid Search Algorithm for Evasive Targets (CRS-AET). The proposed framework utilizes graph-theoretic modeling to represent spatial-temporal relationships among UAVs, targets, and environmental grids. A directional gradient-based motion prediction (DG-Prediction) method first estimates probable movement areas of dynamic targets within the graph-structured environment. An improved multi-round auction algorithm with graph-based utility propagation (IMRAA) then optimizes UAV resource allocation. Finally, Dubins-Constrained Trajectory Optimization (DC-RTO) is integrated within a distributed model predictive control (DMPC) scheme to ensure kinematic feasibility. Simulation results across three representative scenarios indicate that CRS-AET enables faster target detection, enhanced area coverage, and more efficient coordination than baseline methods. Hardware-in-the-loop (HIL) experiments further confirm the robustness and practical applicability of the framework in realistic operational environments.

The rising scale of distributed training (DT) leads to significant communication overhead in clusters. The vast majority of communication loads are generated based on the all-reduce communication primitive. To optimize the all-reduce communication, in-network aggregation (INA) has emerged as a promising solution. Specifically, INA utilizes programmable switches to aggregate gradients within the network, thereby reducing the communication overhead. Existing DT with INA schemes primarily aggregate gradients from all workers in each iteration. However, the system environment is subject to dynamic resource competition and network congestion. As a result, the completion time of DT depends on the slowest worker, i.e., the straggler. To tackle this challenge, we propose a framework called NAST, the first-of-its-kind framework that combines INA with worker selection to accelerate DT. Specifically, we formulate the problem of INA with worker selection and theoretically analyze the convergence of NAST. Based on the convergence result, we design an efficient algorithm to dynamically determine the number of participating workers in each iteration. Then, NAST selects participants for INA based on the arrival order of workers’ gradients. We implement NAST on a testbed comprising 2 Intel Tofino switches and 9 machines. Experimental results demonstrate that NAST reduces the training time by 16.3%-29.8% and traffic consumption by 25.8%-42.4% compared to the state-of-the-art solutions.

This paper is concerned with the near-field channel estimation (CE) in extremely large-scale multi-input multi-output (XL-MIMO) systems under spatial non-stationarity (SnS). To this aim, we first analyze the channel characteristics in the systems and design a structured SnS-aware mask matrix, which reveals the relationship between the angular-domain block sparsity of the channel and the SnS structure. Inspired by the relationship, we establish a near-field channel model for the XL-MIMO systems under SnS. Second, we propose a model-data hybrid driven approach, termed SBL4CE-Net, to estimate the near-field channel. SBL4CE-Net unfolds the block sparse Bayesian learning (BSBL) algorithm into a multilayer solution framework and designs customized neural networks (NN) based on parameter features of BSBL to learn its hyperparameters. Simulation results demonstrate that SBL4CE-Net achieves good normalized mean square error (NMSE) performance, demonstrating a balance between interpretability and data-driven adaptability for SnS affected near-field channels.

This paper presents a data-efficient Koopman-based tracking model predictive control (MPC) scheme for nonlinear systems. Linear surrogate models are constructed online via a kernel extended dynamic mode decomposition (EDMD) framework in a reproducing kernel Hilbert space (RKHS), for which a proportional approximation error bound is established. The proposed online mechanism restricts model learning to a one-dimensional system trajectory, thereby improving data efficiency by avoiding the incorporation of redundant data. Despite the online construction of the surrogate models, the proposed scheme remains computationally tractable and suitable for real-time implementation. Furthermore, the practical exponential stability of the optimal reachable equilibrium (ORE) associated with a given reference signal is rigorously established. The proposed method is further evaluated via a numerical example.

High altitude platform (HAP) millimeter wave (mmWave) links are sensitive to wind-induced attitude shaking. Small roll, pitch, or yaw tilts will misalign narrow communication beams and markedly reduce the gain of the array antenna. In this paper, we propose a vision-augmented large language model (VA-LLM) that forecasts short-horizon attitudes and pre-steers communication beams before misalignment occurs. Specifically, we design a five-channel vision module to render multivariate flight data into pseudo-RGB images with reversible instance normalization (RevIN) and backbone-aligned normalization. We design a learned cross-variable attention (CVA) branch to condense intra-timestep channel relations into a compact token set. Concurrently, we explore a time-series-aware prompt-as-prefix (PaP) to inject frequency, amplitude, and top- $K$-based periodic statistics from the input flight-data window into the token set. After fusing vision-augmented, learned, and text tokens, a frozen LLM refines the representation; a lightweight temporal mixer with axis-wise heads directly regresses future roll/pitch/yaw, which are mapped to continuous steering angles for the array antenna. On real flight test data, VA-LLM improves the average signal-to-noise-ratio (SNR)-type array-gain ratio by 6.56 % over the baseline, achieves a 7.31 % gain over the baseline at the 12step horizon, and yields up to 14.10 % gains over the ablations.

Evolutionary dynamic optimization (EDO) commonly assumes that environmental changes can be detected from fitness variations and handled through random re-initialization, historical solutions, or learned transition patterns. Online calibration of black-box simulators introduces a different setting, where the dynamic objective is induced by sequential observations and a changing calibration window, rather than being controlled by explicit variables. Fitness variations therefore cannot be directly attributed to regime changes, while the unknown relationship between successive regimes limits conventional adaptation. We formulate this setting as an observation-driven dynamic optimization problem and propose PosEDO, which augments fitness-based EDO with an observation-conditioned parameter-space signal. PosEDO learns this signal online as a posterior distribution over simulator parameters from parameter-trajectory pairs generated during evolutionary evaluation, using posterior shifts for change detection and posterior samples for population adaptation. The new evaluation records are further utilized for online posterior updating without additional simulator calls. Experiments on nonstationary economic and financial simulators show that PosEDO improves calibration accuracy, optimization performance, and change-detection quality over representative EDO baselines.

This paper proposes a self-triggered distributed model predictive control (ST-DMPC) scheme with adaptive prediction horizon for nonlinear multiagent systems (MASs) subject to additive disturbances and asynchronous communication. To address disturbances, we construct a time-varying contraction constraint that explicitly compensates for disturbance-induced deviations; the corresponding deviation set is estimated with reduced conservatism via the Gronwall Lemma. To balance performance and efficiency, a self-triggered mechanism is devised to schedule aperiodic optimizations, coupled with an adaptive horizon strategy that dynamically reduces computational complexity as the system converges. Furthermore, a lengthened sequence transmission strategy is introduced to actively compensate for information lags caused by asynchronous communication. Rigorous theoretical analysis guarantees the recursive feasibility and asymptotic stability of the closed-loop system. Simulation results show the effectiveness of the proposed method.

This paper proposes a consistency-model-based channel estimation algorithm for multiple-input multiple-output (MIMO) systems. The proposed algorithm employs a consistency model (CM) to learn the angle-domain channel distribution and uses the trained CM as a plug-and-play (PnP) generative prior for MIMO channel estimation. The proposed algorithm alternates between a pilot-observation-based data-consistency update and a CM-prior-based denoising update. In addition, the proposed algorithm adaptively selects the penalty parameter according to residual energy and residual whiteness, and adjusts the CM denoising level according to the observed signal-to-noise ratio (SNR), thereby avoiding the performance degradation caused by fixed inference schedules under varying observation conditions. Simulation results show that the proposed algorithm not only reduces the number of inference steps by 50

High-fidelity, scalable market simulation is a key instrument for mechanism evaluation, stress testing, and counterfactual policy analysis. Yet existing simulators rarely achieve mechanism fidelity beyond single-asset intraday settings, microstructure fidelity against historical limit order books (LOB), and computational tractability at market scale in a single system. This paper presents EvoMarket, a discrete-event, multi-agent financial market simulator designed for intervention-oriented experiments in multi-asset and cross-day environments. EvoMarket couples a high-throughput execution core (optimized LOB data structures, hierarchical scheduling under propagation delays, and asynchronous per-asset matching) with explicit institutional mechanisms (market calendars, opening call auctions, price limits, and T+1 settlement). To avoid expensive black-box calibration, EvoMarket introduces an Oracle-guided in-run self-calibration mechanism that interprets microstructure discrepancy as missing order flow and synthesizes corrective orders at recording checkpoints. Experiments on China A-share order-flow and LOB data show close replay alignment over five trading days, fidelity gains from budgeted in-run calibration across depth levels, broad agent order-space coverage, and scalable performance under increasing input order rates and market breadth. We further demonstrate cross-asset linkage and event-study style intervention evaluation that produces structured dependence and interpretable event-time responses.

The emerging high-altitude platform (HAP) networks are envisioned as critical components in space-air-ground integrated networks. This paper investigates the uplink channel estimation for large-scale reconfigurable intelligent surface (RIS)-aided HAP networks. To overcome the HAP shaking effect and high computational overhead of massive passive arrays,we propose a shaking-aware fast three-stage channel estimation (SA-FTCE) algorithm in the angular domain, tailored for uniform planar arrays (UPAs). SA-FTCE achieves a computationally efficient estimate by progressively pruning the angular channel matrix to a lower dimension by eliminating inactive azimuth and elevation angular regions. Specifically, in Stage 1, we derive the angle-of-arrival (AoA) interval for the RIS-HAP link through the spatial relationship between the AoA variation and HAP attitude shaking, and introduce a shaking-aware AoA search for initial pruning. In Stage 2, a novel Kronecker variational Bayesian inference (Kronecker-VBI) algorithm is proposed for the low-complexity detection of the effective angular region (EFAR) for further pruning. Finally, the channel estimation is efficiently obtained by a VBI based estimator within the drastically reduced angular space. The simulation results show that the proposed SA-FTCE scheme is faster than its counterparts and achieves comparable estimation accuracy.

The configuration of physical parameterization schemes in Numerical Weather Prediction (NWP) models plays a critical role in determining the accuracy of the forecast. However, existing parameter calibration methods typically treat each calibration task as an isolated optimization problem. This approach suffers from prohibitive computational costs and necessitates performing iterative searches from scratch for each task, leading to low efficiency in sequential calibration scenarios. To address this issue, we propose the SEquential Evolutionary Transfer Optimization (SEETO) algorithm driven by the representations of the meteorological state. First, to accurately measure the physical similarity between calibration tasks, a meteorological state representation extractor is introduced to map high-dimensional meteorological fields into latent representations. Second, given the similarity in the latent space, a bi-level adaptive knowledge transfer mechanism is designed. At the solution level, superior populations from similar historical tasks are reused to achieve a "warm start" for optimization. At the model level, an ensemble surrogate model based on source task data is constructed to assist the search, employing an adaptive weighting mechanism to dynamically balance the contributions of source domain knowledge and target domain data. Extensive experiments across 10 distinct calibration tasks, which span varying source-target similarities, highlight SEETO's superior efficiency. Under a strict budget of 20 expensive evaluations, SEETO achieves a 6

Small changes in high-altitude platform (HAP) attitude can cause significant deviations in HAP downlink beam directions, thereby severely degrading HAP downlink communication performance. In this paper, we develop a multimodal large language model (LLM) enabled beamforming framework to achieve robust HAP downlink communications. Specifically, we design a vision-language LLM (VL-LLM) that learns from multivariate flight telemetry to forecast short-term HAP attitudes under platform shaking and support delay-aware proactive beam steering. We design an offline forecast-error calibration procedure to obtain upper bounds on forecast errors and improve the reliability of proactive analog beam steering. Based on the attitude forecasts, we proactively update the analog beamformer and propose a QoS-driven beamforming and admission method with a lightweight feasibility-enforcement step to satisfy instantaneous transmit-power and QoS requirements. Simulation results indicate that the designed VL-LLM can accurately capture changes in the HAP attitude and the proposed beamforming method achieves a 22.1% higher user service ratio and a 12.5% higher sum-rate than representative baselines. The measured mean and p99 computational latencies are 36.24 ms and 40.13 ms, respectively, supporting low-latency online implementation.

This article presents a safe reinforcement learning (RL) framework for nonlinear multiagent systems (MASs) based on min-max distributed model predictive control (DMPC). The proposed method employs min-max DMPC as a robust baseline to generate control strategies, optimizing closed-loop performance in the presence of disturbances while ensuring interpretability and safety. To mitigate the conservatism inherent in traditional robust DMPC due to its reliance on precise models and fixed disturbance bounds, safe RL is introduced to adaptively update the controller parameters and disturbance sets online. The proposed parameter update mechanism of safe RL formally guarantees the recursive feasibility of the DMPC algorithm during the learning process. Furthermore, theoretical analyses of closed-loop stability are provided. The effectiveness and scalability of the proposed method are validated through two simulation examples.

Dynamic multi-objective optimization requires continuous tracking of moving Pareto fronts. Existing methods struggle with irregular mutations and data sparsity, primarily facing three challenges: the non-linear coupling of dynamic modes, negative transfer from outdated historical data, and the cold-start problem during environmental switches. To address these issues, this paper proposes a decoupled basis-vector-driven generative framework (DB-GEN). First, to resolve non-linear coupling, the framework employs the discrete wavelet transform to separate evolutionary trajectories into low-frequency trends and high-frequency details. Second, to mitigate negative transfer, it learns transferable basis vectors via sparse dictionary learning rather than directly memorizing historical instances. Recomposing these bases under a topology-aware contrastive constraint constructs a structured latent manifold. Finally, to overcome the cold-start problem, a surrogate-assisted search paradigm samples initial populations from this manifold. Pre-trained on 120 million solutions, DB-GEN performs direct online inference without retraining or fine-tuning. This zero-shot generation process executes in milliseconds, requiring approximately 0.2 seconds per environmental change. Experimental results demonstrate that DB-GEN improves tracking accuracy across various dynamic benchmarks compared to existing algorithms.