Drones, Volume 10, Issue 9 (September 2026) – 78 articles

In sudden disaster scenarios, the rapid strategy-driven task-chain construction of rescue task chains in UAV-assisted heterogeneous emergency rescue systems is critical for an effective emergency response. Such systems usually involve multiple types of rescue entities, such as reconnaissance UAVs, delivery UAVs, decision centers, and other emergency resources, which provide four-dimensional capabilities including reconnaissance, intelligence analysis, command and control, and rescue disposal. However, existing manual planning and heuristic optimization methods often suffer from low online decision efficiency in large-scale disaster scenarios and rarely examine how different task-chain construction strategies affect task-chain construction. To address this problem, this paper studies a strategy-aware rescue task-chain construction problem for UAV-assisted heterogeneous emergency rescue systems, in which heterogeneous rescue entities are mapped to reconnaissance, intelligence-analysis, command, and disposal nodes according to their four-dimensional capabilities under capability, distance, cost, availability, and strategy constraints. The problem is formulated as a constrained combinatorial optimization problem, and eight task-chain construction strategies are designed from three dimensions: command mode, information-sharing scope, and disposal mode. To solve this problem, a deep reinforcement learning framework integrating a pointer network and entropy-regularized advantage actor–critic (A2C) is proposed. The pointer network generates variable-length entity-selection sequences for task chains, while entropy-regularized advantage actor–critic (A2C) optimizes the selection policy using an entropy-regularized objective that considers success probability, execution cost, and resource utilization. Experiments across 72 scenarios show that the proposed method outperforms the genetic algorithm in terms of timeliness and overall multi-indicator performance, especially in large-scale rescue scenarios. The results further indicate that distributed command strategies achieve a more balanced and robust performance, providing practical guidance for strategy selection and resource scheduling in UAV-assisted heterogeneous emergency rescue. Specifically, compared with the genetic algorithm under the full-scale 72-scenario evaluation, the proposed method improves the cost-control indicator E2 by 51.2% and the topological indicators E4, E5, E6, and E7 by 86.8%, 346.5%, 375.7%, and 421.5%, respectively, all while reducing the online decision time from 36.5 s to 3.0 s (a 91.8% speedup). The genetic algorithm still outperforms the proposed method on the success-rate indicator E1 and entity-utilization indicator E3, which is discussed as a limitation of the proposed method in the discussion of applicability. Full article

Human detection in drone imagery is an important capability for applications that require information on the location and number of people, including search and rescue, urban pedestrian-flow analysis, crowd-density assessment, and public-space safety management. However, in oblique drone imagery, acquiring and annotating sufficient real data remains difficult because of crowd density, privacy constraints, background clutter, and position-dependent scale variation. This study proposes a copy-and-paste-based synthetic data augmentation procedure for imagery acquired using a DJI Phantom 4 RTK under a 30° camera pitch angle and a 30 m flight altitude, following DJI search-and-rescue observation guidelines. The proposed procedure combines human-free drone background images with a public full-body human dataset and incorporates perspective-based geometry-aware scaling, context-aware placement, and density-matched chip selection based on real person-count distributions. Three YOLO-family detectors (YOLOv8x, YOLO11x, and YOLO26x) and the Transformer-based RT-DETR-X were evaluated at three real-data proportions (10%, 25%, and 100%). Across these twelve combinations, four augmentation strategies (random, geometry-, context-, and combined geometry- and context-aware augmentation) were compared under the same 50% synthetic ratio. For the YOLO family, geometry-aware augmentation improved mAP50-95 in eight of the nine combinations and achieved a mean gain of +0.64 percentage points over the real-only baseline, with the largest gain of +1.99 percentage points under the YOLO11x 25% real-data condition. For RT-DETR-X, the corresponding changes were +0.23, −1.83, and −0.15 percentage points at 10%, 25%, and 100% real data, respectively. These results suggest that, under the acquisition conditions evaluated in this study, aligning pasted person size with the image-position-dependent perspective scale can improve the effectiveness of synthetic augmentation when the chip-level person-count distribution is controlled to match that of the real training data. Full article

Recent autonomous exploration systems for unmanned platforms have improved motion efficiency, trajectory smoothness, and replanning continuity, but thorough coverage remains difficult in confined environments with short branches and partially resolved connectors. We present TOVEX, a completion-oriented framework that couples unresolved Euclidean signed distance field (ESDF) voxels with a sparse topological voxel-sphere (TVS) graph. Evaluation includes a bridge simulation, five-run maze comparisons with the baseline systems EDEN, GBPlanner, and TARE, component ablation, runtime and memory profiling, a cross-structure cave-scene stress test, and real-world experiments with an unmanned aerial vehicle (UAV) in an underground garage and an unmanned ground vehicle (UGV) in a warehouse. The bridge test shows coherent mapping and continuous velocity profiles. In the maze, TOVEX achieves the highest median final coverage of 96.2%, exceeding TARE by 5.2 percentage points, while TARE remains stronger in early coverage and motion efficiency. The geometry-derived terminal-tail analysis confirms that TOVEX achieves the most complete residual-passage resolution, while removing short-branch priority produces the largest loss. At approximately 1100 TVS vertices, median target-selection latency is 0.89 ms and planner proportional set size (PSS) is 52.3 MiB. Field trials demonstrate the same gain-bearing decision layer on aerial and ground platforms. TOVEX is well suited to unmanned-platform missions that prioritize resolving accessible residual passages. Full article

Precision Approach Path Indicator (PAPI) systems are critical visual aids for supporting flight crews during the final approach phase by providing visual information on the aircraft’s vertical position relative to the desired glide path. Their correct operation and angular setting are therefore essential for maintaining operational safety at aerodromes. This study develops and field-evaluates a deterministic UAS-based workflow for automatic PAPI light localization, RED/WHITE/UNDEFINED state classification, red–white transition detection, and angular reconstruction from geotagged UAS observations. Data were acquired at two Colombian aerodromes, Perales Airport (SKIB) and Flaminio Suárez Camacho Aerodrome (SKGY), using a DJI Matrice 400 equipped with Zenmuse P1 and H30T optical payloads under manual vertical flight profiles and two illumination conditions. The proposed algorithm integrates luminance-based saliency extraction, four-light geometric validation, localized ROI-based chromatic classification, and RED → WHITE transition-event detection to estimate the transition angle of each PAPI unit from geotagged UAS observations. Results differed between the two evaluated site–payload configurations. In the SKGY–H30T configuration, the use of zoom and background-attenuation settings facilitated target isolation, and all 255 processed images were correctly classified during manual verification. In the SKIB–P1 configuration, 96.57% image-level classification accuracy was obtained, with the observed misclassifications mainly associated with city lights and luminous halos between adjacent PAPI units. Angular agreement with theodolite measurements also differed between the two datasets. Because each optical payload was evaluated at a different aerodrome, these differences cannot be attributed independently to payload characteristics; they represent the combined response of the payload, acquisition settings, background complexity, illumination, and site-specific conditions. Full article

Accurate identification of pavement distress in UAV imagery is crucial for road safety but is hindered by scale disparities, background noise, and isotropic extraction constraints that neglect highly directional distress textures. This paper proposes MAG-YOLO, a lightweight detection network driven by multi-scale and anisotropic gating perception. The architecture integrates three core innovations: a Re-parameterized Multi-scale Attention (RMA) module, which employs multi-branch depthwise convolutions to enhance fine-grained texture capture while mitigating parameter redundancy; an Anisotropic Gating Feature Pyramid Network (AGFPN), which incorporates orthogonal feature decomposition and dynamic gating mechanisms to extract multi-scale topological features and suppress environmental noise; and a Cross-stage Multi-scale Gated Linear Attention (CMGLA) module designed for global context aggregation and local feature recalibration. Extensive experiments on the RDD2022_China_Drone and UAPD datasets demonstrate that MAG-YOLO outperforms the YOLOv11n baseline by 5.4% and 7.0% in mAP50, respectively, while reducing the parameter count by 29.1%. Deployment tests on the RK3588 edge platform further confirm that MAG-YOLO achieves a favorable balance between accuracy and real-time efficiency. This work provides a robust and deployment-ready solution for UAV-based pavement distress detection, supporting automated road inspection and intelligent pavement maintenance. Full article

Reliable local three-dimensional (3D) wind estimates are important for fixed-wing UAV flight under wind disturbances, but low-cost platforms lack direct 3D flow sensing. We propose PIRNN-AKF, which combines a physics-informed gated recurrent unit (PI-GRU) with a measurement-noise-adaptive Kalman smoother. PI-GRU embeds the wind triangle and attitude rotations in an airspeed-closure loss and predicts covariance scales for adaptive fusion. In PX4/JSBSim simulations with Dryden turbulence, five-seed PI-GRU attains a 3D RMSE of $0.208\pm 0.003$ m/s and a direction MAE of 3.22° on Test-ID. PIRNN-AKF attains $0.544\pm 0.007$ m/s and 1.97° on Test-OOD, reducing RMSE by 37.4% relative to Vanilla GRU; KalmanNet yields lower OOD RMSE but weaker ID accuracy and direction estimation. Session-aware evaluation shows that the AKF reduces OOD jitter by 24.8% while preserving RMSE. In ten frozen HITL sessions of the final 41-input, 100-step model, ID/OOD RMSEs are $0.569\pm 0.054$/$0.832\pm 0.318$ m/s, the companion-side processing $p_{95}$ is $20.4\pm 0.5$ ms, and the achieved loop rate is $83.2\pm 1.6$ Hz with $5.6\pm 0.7\%$ deadline misses. These results support 50 Hz feasibility under simulation truth; real-flight accuracy and closed-loop benefits remain untested. Full article

Most unoccupied aerial systems (UAS) research in cetacean monitoring has shown that drones can acquire high-resolution imagery, but the conditions under which such imagery becomes auditable ecological evidence remain less clearly defined. A structured narrative synthesis (evidence map) was conducted using 242 Web of Science Core Collection records retained from a final export dated 23 June 2026, with search transparency supported by index specification, sentinel-paper recall checking, and predefined evidence and validation maturity coding. Five application domains were identified: morphometrics and health assessment (79 records, 32.6%), behavioral monitoring (76, 31.4%), individual identification (34, 14.0%), abundance and distribution monitoring (33, 13.6%), and multimodal or operational applications (20, 8.3%). Across these domains, the field is shifting from opportunistic visual documentation toward calibrated photogrammetry, computer-vision-assisted detection and re-identification, behavior quantification, and targeted health or molecular sampling. However, ecological inference remains constrained by availability bias, perception bias, group-size error, measurement uncertainty, algorithmic-transfer bias, and disturbance-induced bias. UAS monitoring is therefore evaluated as an observation-to-inference workflow linking platform design, sensor geometry, image preprocessing, annotation, model validation, and uncertainty propagation to management-relevant cetacean evidence. Full article

Unmanned aerial vehicles (UAVs) are increasingly used for airflow measurements because they enable sensing where traditional instrumentation is difficult to deploy. However, most existing approaches rely on hovering or predefined trajectories, which can introduce aerodynamic disturbances and limit operation in confined environments. This study investigates a semi-Lagrangian measurement concept for miniaturized UAVs in which drift is permitted and explicitly accounted for in the reconstruction. Experiments were conducted using a Crazyflie 2.1+ quadrotor (≈35 g total mass, 92 mm footprint) equipped with a differential pressure sensor to measure relative airflow. A stereo-vision system measured the UAV ground-relative motion. Wind velocity was reconstructed by combining the UAV drift velocity with the relative airflow. Laboratory experiments over 2–$7\mathrm{m}/\mathrm{s}$ show that the combined reconstruction improves the wind estimate compared with drift velocity alone. The propagated standard uncertainty is approximately $0.88$–$1.28\mathrm{m}/\mathrm{s}$, and the proposed platform reduces UAV mass by nearly a factor of 20 compared with previously reported setups. However, signal filtering and the time-averaged reference field limit the assessment of instantaneous measurement accuracy. These results demonstrate the feasibility of semi-Lagrangian airflow sensing with very small UAVs while identifying aerodynamic interference and the onboard pressure measurement as the main limitations. Full article

Radio frequency (RF)-based drone identification enables passive low-altitude sensing, but its performance degrades under low signal-to-noise ratio (SNR) and long-range propagation. Existing deep models mainly learn spectrogram amplitude textures and underuse the second-order structure and cross-channel correlations of multi-channel RF signals. We propose the Multi-resolution Riemannian-Spherical Network (MRS-Net), a robust identification framework centered on Riemannian Structure Preservation (RSP). RSP derives noise-referenced Riemannian distance, local geometric variation, log-determinant, and multi-scale statistics from local symmetric positive definite (SPD) covariance matrices. Pairwise similarity alignment transfers these structural relationships into the convolutional neural network (CNN) embedding space as a training-stage teacher signal. RSP is removed at inference and therefore adds no online manifold computation. Multi-resolution short-time Fourier transform (STFT) representations provide complementary weak-signal observations, while CosFace enlarges inter-class angular margins. Experiments on DroneRFa and Noisy Drone RF evaluate propagation attenuation and additive noise, respectively. With four-channel DroneRFa input, MRS-Net achieves 90.30% overall balanced accuracy. Compared with SR-CNN-4ch, it improves the 80–150 m result from 61.30% to 78.30%; compared with MR-Spherical-4ch, it reduces the 5-fold standard deviation from 16.78% to 1.46%. Ablations show that RSP is most effective when cross-channel covariance is informative and the backbone preserves local time-frequency structure. Full article

Natural-language interfaces now influence mission generation in UAV planning pipelines, where a feasible plan may still rest on unsupported completions, unauthorized relaxations, or consequence-changing repairs. We present EAMSR, an evidence-carrying Mission Admission Contract framework that treats a submitted mission as an admission contract, not a text-to-specification output. A large language model proposes candidate clauses and bounded refinements but does not decide admission; the governance layer links hard clauses to evidence and authorized sources, isolates unsupported semantic increments, checks mission-consequence compatibility, and accepts only candidates with a backend task-level witness. The final decision (ADMIT, CLARIFY, or REJECT) carries an audit trail linking language anchors to governance and backend outcomes; backend search failures that do not establish model-level infeasibility return CLARIFY, not REJECT. On EAMSR-Bench (120 tasks, six scenarios, six risk types), EAMSR achieves 100.0% binary admission accuracy ($Ac{c}_{\mathrm{bin}}$) and 93.3% three-class accuracy ($Ac{c}_{\mathrm{adm}}$), with no unwarranted ADMIT decisions among 78 non-admissible cases (0/78; 95% Clopper–Pearson upper bound, 3.8%); residual errors lie on the CLARIFY–REJECT boundary. On an independent external test set (48 tasks), EAMSR achieves 89.6% three-class and 100.0% binary accuracy with no unwarranted admissions. A deterministic non-LLM variant using the same governance and backend layers also attains zero unwarranted admissions but lower accuracy, indicating that LLM generation improves semantic coverage while governance and backend checks determine admission. The evaluation is limited to pre-execution mission admission under stated assumptions and does not constitute flight-safety assurance. Full article

Cooperative path planning in complex coastal mountainous environments near the shore is a knowledge-intensive engineering task crucial for demanding multi-UAV operations such as disaster rescue, environmental monitoring, and coastal emergency response. However, the complex coupling constraints in 3D space make it a challenging high-dimensional optimization problem where traditional metaheuristics often suffer from undirected blind exploration and invalid trial and error. To overcome these limitations, we propose the deep neural network-guided phototropic growth algorithm (DNN-PGA), an engineering informatics framework that deeply couples explicit spatial knowledge representation with traditional heuristic search mechanisms. Specifically, we implement a spatial prior generation method to transform unstructured 3D environmental data into structured knowledge distributions. We utilize deep neural networks to extract spatial heatmap priors from multi-channel voxel representations, serving as a formal representation of potential high-quality path regions. Subsequently, the core search operator of PGA is reconstructed by deploying these explicit spatial priors as structural constraints to strictly guide engineering decision-making. Within the DNN-PGA, a neural network-guided initialization mechanism ensures efficient spatial focusing of the initial population. Secondly, a neural control golden-sine strategy was designed to enable dynamic regulation of local exploitation intensity, and a neural limited lens opposition-based learning mechanism was developed to accurately guide individuals out of local optima using learned spatial knowledge. Extensive experiments confirm that the DNN-PGA achieves an average reduction of 29.49% in mean path cost compared with the best-performing competing algorithm across all scenarios under complex operational conditions. Ultimately, the DNN-PGA successfully establishes a highly robust, knowledge-driven practical solution for demanding 3D multi-UAV cooperative path planning tasks in coastal mountainous environments. Full article

Aquatic vegetation is critical for maintaining water quality, dissolved oxygen levels, and suitable habitats for crabs in aquaculture ponds. However, traditional manual surveys for monitoring aquatic plant growth and identifying invasive weeds are inefficient, labor-intensive, and prone to subjective errors. To address these limitations, this study proposes an integrated framework combining unmanned aerial vehicles (UAVs), remote sensing, and deep learning (DL) for pixel-level monitoring of aquatic plants in crab ponds. High-resolution UAV RGB and multispectral (MS) images were collected by a drone in Changxing County, Zhejiang Province, China. A specialized dataset was constructed, including cultivated Elodea canadensis and Hydrilla verticillata and two dominant invasive weed species (Alternanthera philoxeroides and Lemna minor). Next, an improved U2Net was designed to fuse the RGB-MS images and provide more detailed information. We also propose a SAM-based model, PondSAM, which integrates three modules: (1) a multiscale vision transformer encoder, (2) a self-generated prompt module, and (3) a multilayer aggregation decoder. The Pond-SAM results are relatively better than those of other state-of-the-art methods (e.g., SAM), achieving a Dice score of 0.865, an mIoU score of 0.858, and a recall score of 0.907. Additionally, the proposed method has been tested on the dataset with different light intensities and training ratios, and its reliability was verified. This study confirms that UAV remote sensing combined with deep learning provides a rapid, nondestructive, large-scale solution for aquatic plant monitoring and invasive weed identification in crab ponds, supporting precision aquaculture and ecological regulation. Full article

With the widespread application of unmanned aerial vehicles (UAVs) across various domains, the reliability and lifespan prediction of their core power units—the engines—has become a critical research focus. This study addresses the degradation characteristics of UAV engines under complex operating conditions, including high temperature, high pressure, high rotational speed, and severe vibration, and proposes a remaining useful life (RUL) prediction model based on multi-sensor data fusion. First, a multi-sensor data acquisition platform for UAV engines was established, enabling synchronized collection of multi-dimensional parameters across the entire life cycle, including thrust, torque, temperature, vibration, current, and voltage. Subsequently, a multi-sensor fusion-based RUL prediction model for UAV engines was developed, employing an attention-guided multi-scale residual convolution module to extract local multi-scale degradation features, and integrating a residual-attention Transformer to enhance the modeling of long-sequence dependencies. Experimental results demonstrate that the proposed method outperforms conventional CNN, RNN, and fusion models in terms of RMSE, R², and Score metrics, significantly improving the accuracy and training stability of UAV engine lifespan prediction. This study provides both data support and methodological innovation for predictive maintenance of UAV engines, contributing to enhanced flight safety and mission assurance. Full article

Large unmanned cargo airships may support heavy-lift logistics in regions without runway infrastructure, but payload release produces a rapid buoyancy surplus and changes the vehicle mass properties. This study develops a simulation framework coupling six-degree-of-freedom variable-property flight dynamics, an active seawater ballast system, and constrained ballast-flow allocation. The dynamics are referenced to a fixed body origin and retain the spatial-mass-matrix derivative and declared exchange-momentum wrench. A one-dimensional Method-of-Characteristics (MOC) solution provides a numerical reference for the reduced line-inertance runtime model. Fitting yields $L_{\mathrm{eff}}=55.240$ m and a 15.3% closure-interval normalized root-mean-square error (NRMSE), providing cross-model verification rather than experimental validation. Under the nominal 1201 s mission, the variable-property-aware case satisfies the predeclared criteria with a final-altitude error of 2.985 m and a steady-climb pitch RMS error of ${0.165}^{\circ }$. A fair frozen-inertia ablation also passes and produces slightly lower nominal errors (2.806 m and ${0.157}^{\circ }$); hence, nominal superiority is not claimed. Across the five-point pitch-inertia sweep (0.70 to 1.30 times nominal), the variable-property-aware controller keeps the steady-climb pitch RMS error within ${0.137}^{\circ }$–${0.165}^{\circ }$, whereas the frozen-inertia proportional–integral–derivative (PID) controller spans ${0.153}^{\circ }$–${0.230}^{\circ }$; at 1.30 times nominal inertia, the variable-property-aware controller reduces pitch RMS error by 31.8% and settles 12.9 s earlier. All 21 independently rerun cases that jointly scale the nominal pump and valve time constants from 1.0 to 3.0 satisfy the declared criteria. In the $N=20$, $\pm 5\%$ local parameter-dispersion study, all plotted trajectories remain state-bounded, while the wide altitude spread precludes a uniform tracking or reliability claim. The evidence supports numerical feasibility within the explicitly tested nominal, inertia, and actuator-time-constant cases while requiring configuration-specific trim and controller rematching before extrapolation; it does not constitute a reliability probability or global stability proof. Full article

Pan–tilt–zoom (PTZ) cameras are central to visual counter-UAV surveillance: they steer the optical axis to keep a small, agile target centered and adequately resolved, which tightly couples perception with camera control in a closed loop. Learned PTZ controllers face three obstacles: temporal models that treat the observation stream as exogenous and therefore ignore the agent’s own influence on it; loss detection and recovery driven by hand-tuned thresholds with no measurable notion of reliability; and a costly reliance on real flight data and physical hardware for training. We present CMW-Track, which combines an action-conditioned hierarchical Mamba policy that modulates the state-transition operator with the executed PTZ command; an ensemble state predictor that is deliberately lightweight—it predicts only the low-dimensional target state required for PTZ control rather than reconstructing future frames—and is calibrated by split conformal prediction; and Active Uncertainty-Gated Exploration for Recovery (AUGER), a tracking–uncertain–recovery machine gated by conformal-interval violations instead of a tuned confidence threshold. The controller is trained only on procedurally generated trajectories under domain randomization, with no physical PTZ platform in the loop. Evaluated zero-shot in an unseen high-fidelity Unreal Engine 5 environment, CMW-Track attains the highest tracking-success (81.7%) and loss-recovery (85.0%) rates among all evaluated controllers and matches the best completion rate, in real time (median: 2.5 ms per control step). Against five baselines spanning classical, filtered and recurrent-reinforcement-learning control, it improves tracking success by 5.7 pp over the strongest of them in simulation (95% CI: [+4.4, +6.9]), and the margin persists when that baseline is widened to matched policy capacity. The split-conformal bound holds at calibration time, but closed-loop coverage falls 15–21 percentage points below nominal—a direct measurement of exchangeability breakdown that the deliberately redundant trigger absorbs. We therefore report the trigger as calibrated and auditable, not as a deployment-time guarantee. Full article

Low-altitude unmanned aerial vehicles (UAVs) need environmental evidence for perception and navigation, yet dedicated onboard sensing and prior maps may be unavailable. Existing cellular infrastructure offers persistent signals of opportunity, but a passive UAV observes only scalar RSRP, which conflates transmit power, sector response, and environmental loss. We formulate UAV environmental perception as passive, protocol-assisted sensing using non-cooperative commercial cellular downlinks and propose a protocol-anchored reconstruction of a propagation-equivalent virtual radio environment map (vREM). Decodable nominal reference power and path-loss normalization set the power scale, coarse site bearings constrain antenna directions, and multi-altitude UAV trajectories excite height dependence. A hard height-class Beer–Lambert model recovers occupancy support and a height proxy without building geometry at inference. In a controlled multi-altitude simulation based on Xi’an Bell Tower geometry, the method achieved a held-out ${\mathrm{RMSE}}_{\mathrm{RSRP}}$ of $5.24$ dB, compared with $15.70$ dB for NoProtocol; all metrics improved across 200 paired repetitions. A separate ground-based field collection yielded 1518 observations from 82 LTE/NR sources at 41 sites and reduced relative-power mapping RMSE from $7.27$ to $3.41$ dB. The receiver uses public broadcast fields locally, while the commercial transmitters remain outside the sensing and inversion loop. Full article

Low-altitude unmanned aerial vehicles operating in mountainous environments require continuous spatial reference paths that satisfy terrain-clearance, altitude, pitch, and curvature constraints. Conventional sampling-based planners can efficiently discover collision-free paths but often produce geometrically irregular solutions, whereas swarm-based optimizers are sensitive to the quality and feasibility of their initial populations. This study proposes a feasibility-first hybrid reference-path planner, termed RRT*-DLPSO, which combines Rapidly Exploring Random Tree Star initialization with a particle swarm optimizer incorporating Differential Evolution and Lévy-flight mechanisms. RRT* first supplies a terrain-aware initial polyline, which is compressed into curvature-selected control points and represented by a cubic spline. Because compression and interpolation can invalidate a collision-free polyline, every reconstructed path is checked using the full set of hard constraints. DLPSO then refines the path using time-varying learning factors, stage-adaptive differential trial generation, and stagnation-triggered Lévy perturbations. Feasible paths are ranked using a weighted objective that considers path length, altitude, pitch demand, and horizontal curvature. Experiments were conducted in three DEM-based mountainous scenarios using 50 independent runs per method. The comparison includes a complete RRT*-PSO ablation chain and an RRT*-STOMP baseline. RRT*-DLPSO achieved a 100% continuously certified feasibility rate and the lowest mean final objective value in all three scenarios with respect to the adopted spline and interpolated DEM model. Relative to the strongest competing method in each scenario, it reduced the mean objective by 6.4%, 5.1%, and 3.9%, respectively. Compared with DLPSO without RRT* initialization, the corresponding reductions were 13.0%, 19.6%, and 32.9%. Paired Wilcoxon tests with Holm correction confirmed statistically significant improvements over all major competing methods. The method also reached the scenario-specific objective thresholds in 86%, 100%, and 96% of the runs. Sensitivity analyses repeated in two DEM scenarios supported nine control points and an RRT* step-size and rewiring-radius coefficient pair of (0.020, 0.040) among the tested settings. These results support RRT*-DLPSO as an effective offline spatial reference-path planner for known static mountainous terrain represented by DEM data. Full article

Computing services can be extended to infrastructure-limited areas through unmanned aerial vehicle (UAV)-assisted mobile edge computing (MEC). However, service provision continues to be challenging when spatially clustered ground users (GUs) are initially unknown and their intermittent workloads expire within finite validity windows. Therefore, conventional trajectory controllers assuming known GU locations and requests are unsuitable under sensing-limited conditions, where out-of-range workloads remain hidden between visits. Partially observable control has been adopted to address these challenges; however, existing frameworks are insufficient, as the discovery of unknown service regions and the repeated revisits required to re-observe hidden demand are not jointly considered. In this paper, a deadline-paced patrol with goal-conditioned multi-agent reinforcement learning (DPP-GCMARL) framework is proposed. In the proposed framework, a rule-based deadline-paced patrol (DPP) layer converts sensing-derived cell memory into deconflicted target cells using staleness normalized by the collection deadline. The assigned targets are then mapped to continuous movement commands by a learned goal-conditioned movement (GCM) layer. Simulation results show that DPP-GCMARL improves the end-to-end completed-workload ratio compared with multi-agent proximal policy optimization (MAPPO). It also increases exploration-coupled fairness and the worst-cluster collection ratio while reducing the mean travel per UAV. These results substantiate the proposed framework’s potential for deadline-constrained MEC service in infrastructure-limited environments. Full article

Urban low-altitude unmanned aerial vehicle (UAV) planning is inherently spatiotemporal because route feasibility and cost depend on segment arrival times. Search-stage surrogates may therefore favor paths that fail execution-level checks as moving obstacles, temporary no-fly zones, wind-dependent energy use, and building-clearance constraints evolve. We address this search–execution mismatch with the Risk-Aware Artificial Lemming Algorithm (RA-ALA), a three-layer framework governed by a common arrival-time-recursive evaluator. Sequential temporal propagation aligns candidate generation with final assessment, while an energy-weighted A* (Energy-A*) warm start guides continuous waypoint search. The Top-K stage then re-evaluates path variants before feasibility-first selection and conditional recovery. Under prespecified algorithm-specific budgets across 10 High-complexity environments, RA-ALA achieved the highest observed evaluator-feasible rate (24/30, 80.0%), 20 percentage points higher than Energy-A* and space–time Energy-A* (ST-EA*). After Holm adjustment, these contrasts were nonsignificant, while differences against Informed-RRT* and Greedy were supported. Within jointly feasible environments, RA-ALA retained competitive composite scores. Same-cohort descriptive ablation associated Top-K removal with higher composite scores and more infeasible outputs. These results support RA-ALA as a simulation-tested route-generation framework under the modeled constraints, without establishing isolated-operator superiority or real-flight readiness. Vehicle dynamics, sensing, tracking, communications, and flight validation remain outside this scope. Full article

Traditional UAVs often suffer severe structural damage during high-speed collisions. Recent research has explored sensing, control, and design strategies to enhance collision resilience. This work presents fully passive, soft continuum-lattice arms that combine visco-hyperelastic materials with nonlinear lattice geometries to trigger controlled buckling of the beams and absorb impact energy. Visco-hyperelastic parameters are identified from quasi-static and dynamic tensile tests on the material and then implemented in finite element models to optimize lattice porosity and distribution for force mitigation under quasi-static and dynamic loading conditions. Experimental validation of the simulations for the optimized beam configuration shows good agreement, with MAE 6.7% and RMSE 7% in quasi-static loading and MAE 12.1% and RMSE 14% in dynamic loading. Controlled drop tests on a quadrotor prototype demonstrate superior impact energy absorption and force mitigation compared with a bulk material design (up to a 99% increase in specific energy absorption and up to a 65% reduction in peak impact force). Furthermore, the latticed UAV is flown voluntarily and then dropped onto the ground, maintaining low thrust losses of 0.42 ± 0.26% and structural integrity after impact. Full article

Terahertz (THz) integrated sensing and communication (ISAC) is a pivotal paradigm for enabling high-rate, ultra-reliable air-to-ground (A2G) connectivity in sixth-generation (6G) unmanned aerial vehicle (UAV) networks. From a physical propagation and environment-aware transmission perspective, current THz channel models inadequately characterize dynamic obstacle scattering, neglect sensing-communication shared clusters, and lack closed-loop frameworks leveraging sensing to assist communication. To bridge these gaps, this paper proposes a novel 3D non-stationary geometry-based stochastic model (GBSM) for THz UAV A2G MIMO channels operating at 300 GHz. The model explicitly incorporates obstacle-induced scattering with Radar Cross Section (RCS)-dependent properties, employing single-point models for small obstacles and multi-point models for large obstacles to capture multipath structures, Doppler effects, molecular absorption, and dynamic cluster evolution. Furthermore, we develop a shared cluster identification framework utilizing delay-angle similarity metrics combined with the Hungarian algorithm to match background scatterers with target-induced paths. Building upon this physical model, a closed-loop sensing-assisted adaptive transmission scheme is established, integrating obstacle-trajectory-driven Kalman channel prediction, affected subchannel avoidance, and waterfilling power allocation. Simulation results demonstrate that the proposed framework accurately characterizes physical THz propagation environments and significantly improves system capacity and reliability, demonstrating the efficacy of environment-level sensing-assisted communication in dynamic THz UAV scenarios. Full article

This paper addresses low-altitude navigation of multiple UAVs through a shared two-dimensional environment with narrow passages, blocked cells, and predicted moving obstacles while preserving an assigned task order. The planner receives ordered task bundles from an allocator and checks static obstacles, moving-obstacle timing, sampled minimum inter-UAV separation, deadlines, risk budgets, and an energy proxy. It generates risk-weighted candidate path segments, repairs timing conflicts with waits and local detours, verifies service and terminal occupancy, and returns a certificate that records whether a segment is executable, its total travel cost, risk exposure, and energy proxy, or the reason for failure. We compare no feedback, context no-good, typed-failure, quantitative, and combined feedback under medium-load and high-stress test suites. Quantitative feedback lowers risk per completed task in both suites after correction for multiple comparisons. Failure-type feedback adds no detectable benefit, and completion-rate differences do not remain significant after the same correction. Fixed-bundle simulations show that the proposed planner can preserve scheduled executability while reducing threat exposure relative to a spatiotemporal-priority baseline. Single-UAV flights demonstrate waypoint execution, reference tracking, and avoidance of designated regions. Full article

As unmanned aerial vehicle (UAV) missions expand from aerial inspection and environmental sensing to physical interaction and autonomous manipulation, rotary-wing UAVs equipped with flexible manipulators offer a promising platform for contact-rich operations in complex environments. Among these tasks, stable near-ground hovering and the grasping of ground targets are particularly challenging because they involve ground-effect aerodynamics, rigid–flexible coupling, and mode transitions caused by contact and load transfer. This review provides a structured, task-oriented critical synthesis of advances in this interdisciplinary field. First, system configurations are classified by aerial-platform architecture, manipulator type, mounting arrangement, and end-effector design, and the suitability of rigid-link, compliant, continuum, and soft manipulation mechanisms for near-ground grasping is assessed. Next, modeling approaches for rotor ground effect, coupled rigid–flexible dynamics, hybrid contact and load-transfer dynamics, model identification, and model reduction are reviewed. Trajectory planning, coordinated stabilization, impedance control, hybrid force/position control, and switching control are then compared across free flight, contact establishment, and payload-carrying hover. Although the reviewed literature provides a substantial theoretical foundation for aerial manipulation, continuum robotics, and multirotor ground effect, direct evidence remains limited for methods that jointly address near-ground aerodynamics, large flexible deformation, and contact-induced load transfer across the complete near-ground grasping sequence with integrated experimental validation. Based on these evidence gaps, this review identifies multiphysics reduced-order modeling and event-driven hybrid control as author-synthesized directions for future investigation. Full article

The nonlinear nature of quadrotor dynamics requires high-fidelity reconstruction of vehicle position and attitude. In order to mitigate singularities associated with global three-parameter representations—such as Euler angles—the quadrotor attitude is conventionally parameterized via unit quaternions and reconstructed by integrating quaternion differential equations. However, the standard quaternion integration procedure is structure non-preserving. It is based on a set of linear differential equations that subsequently enforces the unitary norm of the quaternion through additional algebraic equations. To this end, the paper presents the utilization of the recently introduced structure-preserving attitude and position update algorithms, based on Lie groups, that overcome the drawbacks of the standard procedures. The numerical test cases involve UAV performing five different maneuvers. It is shown that conventional algorithms may suffer from instabilities and errors in kinematical update of position and attitude, which can be circumvented by using the geometric structure-preserving (Lie group-based) algorithms presented here. Full article

Image disturbances caused by platform ego-motion are coupled with true target motion under the dynamic viewpoints of unmanned ground vehicles (UGVs), resulting in trajectory-prediction drift, association mismatches, and persistent contamination of identity information by unreliable observations. This paper proposes ERM-Track, a disturbance-aware and reliability-guided online multi-object tracking framework. I²DF-Mamba uses a causal dual-stream Mamba encoder to integrate IMU, joint-state, and trajectory histories and predicts a trajectory-specific image-disturbance distribution. The predicted disturbance mean and uncertainty are mapped explicitly from normalized/log-scale disturbance coordinates to the detection-observation space. CF-TUR then estimates observation reliability through reliable–contaminated posterior fusion and causal evidence accumulation and generates separate bounded write gains for the motion state and identity memory. On the 6488-frame sealed holdout set of the additionally annotated CEAR data, ERM-Track obtains 64.34% HOTA, 66.28% AssA, and 73.42% IDF1, with 28 identity switches. Three-seed backbone replacement experiments show that Mamba provides the highest mean HOTA among the evaluated causal encoders. Post hoc isotonic calibration reduces ECE from 0.4752 to 0.0478, with only marginal changes in the tracking metrics. The complete pipeline reaches 69.50 FPS on an RTX 4080 workstation and an onboard mean latency of approximately 29 ms on a Jetson Orin NX, corresponding to a reciprocal processing rate of approximately 34.5 FPS. Full article

To address the challenge of dynamic target allocation in multi-unmanned aerial vehicle (UAV) aerial games, this paper proposes a hierarchical solution method that combines matrix game dimensionality reduction with the Eigen-Crossover Differential Evolution (ECDE) algorithm. Assuming that both parties’ UAV platforms are identical, we first model the aerial game as a two-player zero-sum mixed-strategy matrix game, and introduce target value factors and mutual-confrontation cross term to characterize the coupling between target value and antagonism. Second, by performing fast dimensionality reduction on the utility matrix through convex hull vertex enumeration, we effectively compress the strategy space while preserving the strategic information relevant to the original game equilibrium. Finally, we designed ECDE to find the Nash equilibrium in a dimensionality-reduction game. This algorithm uses elite covariance eigendecomposition to rotate the crossover operation into the natural coordinates of the fitness landscape, and combines parameter adaptation with population size reduction strategies to balance search efficiency and convergence stability. Experimental results show that in the CEC2022 test suite, ECDE achieved the top overall ranking among 13 compared algorithms for both $D=10$ and $D=20$; in the 3v2 adversarial scenario, it achieved an average decision time of only 0.026 s and a Nash equilibrium solution accuracy of the 10⁻¹⁵ order of magnitude, while maintaining efficient solution capabilities across matrices of varying sizes. The proposed method meets real-time decision-making requirements while ensuring solution accuracy, demonstrating engineering value. Full article

Satellite downlinks reliably connect remote unmanned aerial vehicle (UAV) swarms, but broad coverage exposes information-bearing signals to unauthorized receivers. This paper proposes a multi-beam cooperative time-varying directional modulation (TVDM) framework for secure satellite downlinks. Two asymmetrically partitioned subarrays form cooperative beams whose pointing states and beam-dependent symbol mappings are randomly updated at the symbol rate, thereby introducing controlled randomness for physical-layer security. In each interval, the source symbol is mapped to two transmit symbols whose superposition remains in the correct phase-shift-keying decision region at legitimate UAVs, while the equivalent constellation varies at unauthorized locations. A relaxed transparent-transmission constraint based on constructive decision-region margins enables conventional detection without instantaneous TVDM-state estimation. Mutual information (MI) defines the pointwise multi-UAV secrecy capacity (SC), main-lobe insecure area, and sidelobe leakage. A mixed discrete–continuous problem jointly optimizes the trajectory radius, relative pointing phase, mapping parameters, and subarray partition to reduce insecure coverage and sidelobe leakage. A block alternating algorithm combines Monte Carlo MI evaluation, projected Armijo updates, and finite partition search to coordinate continuous and discrete variables. Simulations using representative low-Earth-orbit satellite parameters show that the proposed design reduces the central insecure-interval length by 82.7% relative to conventional beamforming. Full article

This article addresses real-time local conflict resolution for a self-planning UAV operating in a three-dimensional dynamic environment with surrounding UAVs. To this end, we develop an SBA–DWA hybrid planning framework in which an improved sticky bacteria algorithm (SBA) is embedded in the dynamic window approach (DWA) to enhance real-time velocity selection for the self-planning UAV. First, a chemotaxis operator with projection is developed to strictly constrain bacterial positions within the convex dynamic window. Furthermore, an anisotropic Gaussian adhesion potential field is proposed to adaptively guide the current population search using historical optimal velocity commands, achieving cross-step memory transfer. Then, a dynamic pruning mechanism is designed to ensure that historical memory does not lead UAVs into infeasible or hazardous regions. The proposed scheme guarantees that the single-step planning latency satisfies stringent real-time requirements. Comparative simulation results demonstrate that the proposed method reduces path length by approximately 30% and planning time by approximately 31% compared with the standard DWA, while achieving a larger minimum inter-vehicle clearance in dense dynamic scenarios. Full article

In disaster response and other infrastructure-limited settings, UAV-mounted access points can rapidly restore service availability for mobile ground users as demand and fleet availability evolve. Existing single-slot coverage formulations, however, can mask prolonged individual outages and do not jointly represent heterogeneous service priorities, finite battery capacities, and periodic recharging. We study persistent geometricmulti-UAV service coverage, where a user is available for service when it lies inside a UAV footprint. We propose Priority- and Outage-Guided Safe QMIX (POGS-QMIX), a hybrid hierarchical framework in which a centralized online coordinator forms conflict-reduced UAV–user targets from fleet-wide priority and outage information, while parameter-shared QMIX agents independently choose target-conditioned low-level actions. The framework couples class-balanced outage memory, assignment, dense target-progress feedback, and a return-energy action mask. The evaluation includes learning and non-learning baselines, greedy-versus-Hungarian assignment, multi-seed statistics, sensitivity studies, operating-condition studies, and energy-stress tests. In the default scenario, POGS-QMIX obtains high-priority coverage $0.547\pm 0.009$ and maximum high-priority outage $38.0\pm 4.7$ slots over five independent seeds. Full article