Keywords: Networked Swarms, Control Architectures
Abstract: Nonholonomic motion constraints fundamentally limit maneuverability in many aerial robots, particularly fixed-wing vehicles that cannot generate instantaneous lateral motion. These limitations become even more pronounced in the presence of exogenous disturbances. Motivated by these challenges, this paper develops a new distributed control framework for disturbed nonholonomic aerial robots consisting of two integrated steps. Specifically, the first step employs feedback linearization to represent each aerial robot with single-integrator dynamics. The second step introduces a switching distributed control protocol that guarantees finite-time synchronization at a user-defined time horizon T and preserves synchronization for all t >= T. Stability of the proposed protocol is rigorously established using methods from time transformation, input-to-state stability, and Lyapunov analysis. In addition to the theoretical developments, an illustrative numerical example is also provided to demonstrate the effectiveness of the proposed control framework.

Keywords: Path Planning, Navigation, Swarms
Abstract: We address decentralized multi-UAV wildfire monitoring as a long-horizon active state estimation problem under partial observability. UAVs integrate local observations into global belief maps for fire and fuel, while an uncertainty map identifies unobserved regions. We propose an information-driven objective combining uncertainty reduction, fire-front tracking, and fuel-aware exploration using a conservative no-fuel map. A decentralized MCTS planner provides high-quality decisions but is computationally costly online. We distill its behavior into a diffusion-based trajectory policy, amortizing planning into learned conditional sampling of future poses. Experiments over multiple scales and team sizes show that the diffusion policy surpasses random and greedy baselines and closely matches MCTS in estimation accuracy. Its runtime is comparable to MCTS at small scales, but it scales more favorably and delivers a better accuracy–latency trade-off in larger environments.

Keywords: Swarms
Abstract: This paper introduces an innovative Human-Swarm Interaction (HSI) architecture leveraging a Brain-Computer Interface (BCI) to evaluate operator trust in real-time during collaborative multi-UAV navigation. While Electroencephalography (EEG)-based trust detection is well-documented, its implementation remains predominantly restricted to offline assessments. Addressing the critical need for online, adaptive control in aerial robotics, we propose a k-Nearest Neighbors (k-NN) classifier that differentiates trusting from distrusting cognitive states with over 90% accuracy. Integrated into a closed-loop BCI, this model dynamically adjusts aerial swarm formations. Experimental validations demonstrate the system's capacity to swiftly identify operator distrust, prompted by sub-optimal UAV behaviors, and autonomously revert the swarm to a stable, trusted configuration. These findings highlight that higher-order human factors, particularly trust, can effectively govern adaptive controllers in unmanned aerial systems, facilitating hands-free operations for complex missions.

Keywords: Swarms, Autonomy, Air Vehicle Operations
Abstract: We present Swarm-Steward, a platform-agnostic system for natural-language multi-robot control that enables a non-expert operator to coordinate many drones by creating and commanding groups, bridging high-level intent to reliable, low-level execution. Swarm-Steward uses a hierarchical LLM-based multi-agent design where planning and context gathering are separated from actuation: specialized agents ground requests in map features and telemetry, while a dedicated action-execution stage composes group actions and dispatches only deterministic, schema-constrained commands through a safety gate (e.g., geofencing, altitude and separation limits), with optional operator preview before execution. To keep grounding scalable, Swarm-Steward applies dual retrieval-augmented generation over both map features (Feature RAG) and telemetry variables (State RAG), injecting only relevant candidates at each step. In scalability tests, Feature RAG remains effective with up to 10,000 features, and end-to-end sessions from 5 to 500 drones maintain near-constant LLM token cost with 92.9% task success across 280 prompts. The observed failures arise from prompt-specific scaling limits at larger swarm sizes, rather than from incorrect coordinator planning. Finally, we validate sim-to-real transfer by executing the same mission script in simulation and on a real-world DJI Mini 4 Pro swarm, demonstrating consistent behavior across comparable mission phases without modifying the LLM multi-agent system layer.

Keywords: Networked Swarms, Swarms, Micro- and Mini- UAS
Abstract: This work proposes a distributed formation control strategy for a multi-robot system, addressing local sensing constraints and the rigidity of classic methods. By combining a Voronoi-based controller for separation and a consensus algorithm for cohesion, the proposed approach ensures flocking-inspired behaviors using only the relative positions of close neighbors. Each Voronoi cell is enclosed in a circular convex region, called bubble, enabling distributed Centroidal Voronoi Tessellation for safe separation. The strategy requires no global knowledge of the size of the swarm and autonomously adapts to changes in the number of robots. The stability of the system is theoretically proven through the separation scale principle between Voronoi tessellation computation and formation control. Simulation and experimental results highlight the effectiveness of the proposal on a group of Unmanned Aerial Vehicles.

Keywords: Swarms, Perception and Cognition, Navigation
Abstract: Smoke plumes provide critical information for assessing wildfire behavior, spread dynamics, and atmospheric impact. In order to observe such phenomena, single-UAV observations are often limited by restricted viewpoints and insufficient coverage of evolving plume structures. This paper presents a perception-guided coordination framework for UAV swarms that integrates object detection with adaptive motion primitives to enable structured, multi-view observation of smoke plumes. Upon detecting a target of interest, the swarm autonomously executes a sequence of behaviors: a safe straight-line approach, circular orbiting for various lateral perspectives, and vertical sweeps with altitude-dependent radius scaling to capture elevation-dependent plume morphology. The system is implemented on DJI Mini-3 quadrotors, where a designated leader performs YOLO-based detection and relative target pose estimation, and the follower UAVs maintain a collision-avoiding formation around the estimated target position. Outdoor experiments using a visually distinctive vertical column as a surrogate target demonstrate reliable detection, rapid alignment, and stable coordinated orbiting with inward-facing yaw tracking. This work establishes a foundation for perception-driven adaptive UAV swarm systems to enhance smoke observation and reconstruction strategies in wildfire monitoring scenarios.

Keywords: Autonomy, Path Planning
Abstract: Enabling fixed-wing UAVs to maneuver aggressively through cluttered environments—while respecting stall and actuator limits—remains an active area of research. This paper presents an Integral High-Order Control Barrier Function (IHOCBF) safety filter for a 12-DOF nonlinear fixed-wing model that enables such maneuvers without explicit trajectory planning. By treating actuator deflections as states and optimizing their rates, we obtain a control-affine augmented system that naturally accommodates actuator constraints. The safety filter is posed as a quadratic program that minimally modifies inputs from a nominal controller (in this case an ergodic exploration one) to enforce obstacle avoidance, stall prevention, and geofence limits. Simulations demonstrate emergent aggressive behaviors—including ”bank and yank” coordinated turns, reactive wall avoidance near stall limits, and sustained loitering over multiple regions of interest—with computation times below the simulation timestep.

Keywords: Autonomy, Networked Swarms, Path Planning
Abstract: Wildfires pose an increasing global threat, endangering both human and animal lives. Rapid and coordinated search and rescue (SAR) operations are critical to minimizing casualties in such emergencies. This paper investigates the use of Multi-Agent Reinforcement Learning (MARL) to train autonomous unmanned aerial vehicles (UAVs) capable of cooperative SAR in simulated wildfire environments. The task is modeled as a decentralized partially observable Markov decision process (Dec-POMDP) and trained under a Centralized Training with Decentralized Execution (CTDE) paradigm. Two learning configurations are compared: a single-agent baseline using Proximal Policy Optimization (PPO) and a cooperative multi-agent framework based on Multi-Agent Policy Optimization with Credit Assignment (MA-POCA) incorporating posthumous credit assignment. Training employs a three-stage curriculum to progressively increase environmental complexity and enhance policy generalization. Simulations across one to six UAVs demonstrate that multi-agent coordination significantly improves mission efficiency and consistency. Specifically, teams of four to five UAVs achieved the lowest average completion times while maintaining high stability and reliability across trials. These results confirm that MARL-based cooperative control improves scalability, robustness and overall mission performance in UAV-based SAR operations, especially under optimal team sizing, underscoring the potential of decentralized learning for real-world disaster response scenarios.

Keywords: Autonomy, Path Planning, Simulation
Abstract: This paper presents a method for the autonomous inspection of rotating wind turbines using an unmanned aerial vehicle (UAV). To address the challenges of a dynamic environment, the proposed approach tightly couples LiDAR-based wind turbine pose estimation and blade tracking with illumination-aware path planning. Using onboard GPS, IMU and LiDAR sensors, the proposed approach determines the position of the wind turbine relative to the UAV. A multi-objective optimization framework then generates an inspection path which ensures complete blade coverage while maximizing image quality by accounting for environmental constraints. During the inspection flight, a camera mounted on a gimbal is used together with a range-finder sensor to capture high-resolution blade images. Simultaneously, the blade tracking algorithm ensures the correct association between obtained images and specific blades. The proposed approach is validated in a realistic simulation environment, while the wind turbine pose estimation is further analyzed on real-world aerial datasets of operational wind turbines.

Keywords: Autonomy, Simulation, Multirotor Design and Control
Abstract: Autonomous indoor flight for critical asset inspection presents fundamental challenges in perception, planning, control, and learning. Despite rapid progress, there is still a lack of a compact, active-sensing, open-source platform that is reproducible across simulation and real-world operation. To address this gap, we present AgiPIX, a co-designed open hardware and software platform for indoor aerial autonomy and critical asset inspection. AgiPIX features a compact, hardware-synchronized active-sensing platform with onboard GPU-accelerated compute that is capable of agile flight; a containerized ROS2-based modular autonomy stack; and a photorealistic digital twin of the hardware platform together with a reliable UI. These elements enable rapid iteration via zero-shot transfer of containerized autonomy components between simulation and real flights. We demonstrate trajectory tracking and exploration performance using onboard sensing in industrial indoor environments. All hardware designs, simulation assets, and containerized software are released openly together with documentation.

Keywords: Autonomy, Technology Challenges, Training
Abstract: The conceptual design of fixed-wing unmanned aerial vehicles (UAVs) demands aerodynamic evaluation tools that balance physical fidelity with computational efficiency. This work presents a set of multilayer perceptron (MLP) based reduced-order models (ROMs) for predicting lift and drag coefficients across eight conventional and non-conventional fixed-wing configurations, including canard, tandem, joined-wing, and box-wing layouts. Training data were generated using a hybrid low-fidelity methodology combining the Vortex Lattice Method and empirical drag correlations, targeting small- to medium-scale UAVs operating in the low subsonic regime. The developed ROMs achieve mean R^2 values of 0.9995 and 0.9940 for lift and drag prediction across all evaluated configurations, respectively, while requiring less than 10% of the computational time of the underlying low-fidelity solver. Although predictive fidelity is bounded by the low-fidelity training data, the ROMs provide near real-time aerodynamic evaluation suitable for integration into conceptual design and multidisciplinary design optimization (MDO) workflows. Configurations with strong geometric coupling at wing tips, specifically joined-wing and box-wing layouts, exhibited higher prediction errors, a limitation discussed in the context of VLM solver assumptions.

Keywords: Autonomy, Micro- and Mini- UAS, Path Planning
Abstract: Target localization is a fundamental problem for unmanned aerial vehicle (UAV) missions. While existing bearing-only localization methods achieve high accuracy, they often rely on multi-UAV setups or computationally intensive estimation filters that are not suitable for resource-constrained platforms. This paper proposes a novel, low-complexity geometric localization method specifically designed for a single fixed-wing UAV. The proposed approach estimates the target position by iteratively intersecting measurement triangles derived from bearing observations subject to bounded noise. Furthermore, an adaptive doubling-halving trajectory planning algorithm is employed to guide the UAV toward the target while reducing localization uncertainty and ensuring continuous motion. Extensive simulations demonstrate that the proposed method successfully balances localization accuracy with operational cost under varying bounded-noise levels and desired uncertainty thresholds.

Keywords: Energy Efficient UAS, Multirotor Design and Control, Path Planning
Abstract: Today, unmanned aerial vehicles, especially multicopters, are utilized in environmental monitoring, infrastructure assessment, logistics, and disaster response due to their flexibility, maneuverability, and ability to operate in complex environments. However, autonomous coordination, planning, and control of these systems require accurate yet computationally efficient modeling of the employed vehicles and their capabilities. This paper presents a modeling approach that considers vehicle dynamics and energy consumption. The power train model provides a detailed representation of key components, such as lithium-ion batteries, electronic speed controllers, and brushless direct current motors, validated using real flight data. Since most of the model parameters are directly taken from data sheet information, the proposed model only requires few calibration flights. The proposed energy-aware multicopter model serves as a foundation for planning, control, and coordination of unmanned aircraft systems in dynamic environments.

Keywords: Energy Efficient UAS, Path Planning, UAS Applications
Abstract: Inspired by the flight behavior of albatrosses, dynamic soaring has emerged as a promising technique for energy-efficient flight with unmanned aerial vehicles. However, the current implementation of dynamic soaring is restricted by the minimum wind shear requirement and makes limited use of the harvested energy. In this work, we investigate the combined use of thrust and regeneration in dynamic soaring, which gives UAVs more flexibility in energy management, allowing them to operate in a wider range of wind conditions and with a higher energy-harvesting efficiency. In addition, we explore the effect of drive train losses on the use of thrust and regeneration during dynamic soaring, providing insights into the deployment of such a strategy in practice. The results demonstrate the significant benefits of using both thrust and regeneration during dynamic soaring, especially in conditions with strong wind shear.

Keywords: Energy Efficient UAS, Path Planning, Autonomy
Abstract: Battery-powered multirotor unmanned aerial vehicles (UAVs) can rapidly map unknown environments, but mission performance is often limited by energy rather than geometry alone. Standard exploration policies that optimise for coverage or time can therefore waste energy through manoeuvre-heavy trajectories. In this paper, we address energy-aware autonomous 3D exploration for multirotor UAVs in initially unknown environments. We propose Energy-Aware Autonomous Exploration (EAAE), a modular frontier-based framework that makes energy an explicit decision variable during frontier selection. EAAE clusters frontiers into view-consistent regions, plans dynamically feasible candidate trajectories to the most informative clusters, and predicts their execution energy using an offline power estimation loop. The next target is then selected by minimising predicted trajectory energy while preserving exploration progress through a dual-layer planning architecture for safe execution. We evaluate EAAE in a full exploration pipeline with a rotor-speed-based power model across simulated 3D environments of increasing complexity. Compared to representative distance-based and information gain-based frontier baselines, EAAE consistently reduces total energy consumption while maintaining competitive exploration time and comparable map quality, providing a practical drop-in energy-aware layer for frontier exploration.

Keywords: Payloads
Abstract: This work addresses the altitude tracking control problem for agricultural Unmanned Aerial Vehicles (UAVs) subject to time-varying mass due to liquid payload depletion. The continuous change in system inertia and gravitational forces can significantly degrade the tracking performance of conventional fixed-gain controllers. To ensure robust performance, an adaptive prescribed performance control scheme is proposed for the vertical dynamics of the UAV. By utilizing a logarithmic error transformation, the proposed method guarantees that the altitude tracking error evolves strictly within a user-defined, decaying performance funnel, thereby ensuring the pre-specified transient and steady-state characteristics. An adaptive update law is employed to compensate for parametric uncertainties and the effects of mass variation. Numerical validations, performed in a physics-based environment, demonstrate that the controller achieves precise altitude tracking without violating the user-defined performance constraints, despite significant weight loss.

Keywords: Payloads, Simulation, Training
Abstract: Unmanned Aerial Vehicles (UAVs), particularly multirotor platforms, are widely used in civil and public applications. Many missions require payload (P/L) transportation and release, causing significant variations in the system dynamic. These changes, combined with external disturbances, introduce time-varying dynamics that challenge conventional control methods and often require separate controllers for different flight configurations. This work investigates Deep Reinforcement Learning (DRL) strategies for a hexacopter control under dynamic regime transitions. A robust thrust control strategy is developed using probabilistic actor–critic algorithms, both on-policy (PPO, TRPO, RPO) and off-policy (SAC). The control problem is formulated under partial observability: the policy does not receive explicit information about P/L mass, attachment state, or release timing. The DRL agent implicitly infers dynamic changes from onboard observations. The policy is assessed during two different phases:P/L transport–release P/L and collection–release. To enhance generalization across varying operating conditions, domain randomization and adaptive learning strategies are incorporated during training. The approach is systematically assessed under continuous external disturbances in the form of steady wind. The obtained results show that all policies maintain bounded trajectory tracking despite P/L variations and wind disturbances. Stochastic on-policy methods effectively handle dynamic transitions, with PPO and TRPO achieving the highest tracking accuracy and robustness and ensuring stable performance during regime changes, while RPO and SAC exhibit slower convergence and reduced stability in flight validation, consistent with the characteristics of their respective learning frameworks.

Keywords: UAS Applications, Aerial Robotic Manipulation, Payloads
Abstract: This paper describes the design, development, and validation of a 40 by 40 cm size, 2.2 kg weight autonomous aerial delivery robot intended to conduct intra-logistics operations with light parcels (under 250 grams) that can be safely delivered directly to the people on flight. The aerial robot is fed by power cable to overcome the limitations of batteries while increasing the payload-to-weight ratio, covering the propellers with the carbon fiber frame structure and foam protection to avoid the potential damage in case of collision with the people or the environment. A comparison of the proposed design with respect to our previous prototypes is presented along with the related works to motivate the design choices. The aerial robot implements and compares two onboard localization and mapping methods: RTAB-Map with RGB-D camera and 2D LiDAR, and FAST-LIO2 with 3D LiDAR. The platform is validated in two indoor mockup scenarios executing two different tasks. In the first one, the aerial robot is equipped with a hook for retrieving a parcel from a shelf, whereas in the second scenario a parcel is delivered through the window of a user's home.

Keywords: Security
Abstract: This article introduces a Theory of Mind (ToM) model for proportionality assessment in military operations that captures how civilian and adversary beliefs and behaviors modulate collateral damage and military advantage across alternative CoAs (Courses of Action. To this end, the proportionality assessment is formalized as a dual-lens decision problem in which collateral damage is evaluated under two perspectives. First, a physical harm only lens, and second, an expanded lens that includes psychological harm in addition to physical harm, while preserving a common rule-based proportionality classifier mapping categorical results. Furthermore, to demonstrate and evaluate the model proposed, a counter-UAV (Unmanned Aerial Vehicle) engagement use case is considered where a comparison between kinetic, non-kinetic, delay/reposition, warning, and abort CoAs is conducted. The results show that the ToM model provides a transparent, distributional, and policy-relevant approach to proportionality assessment while capturing behavioral feedback loops and clarifying when assessment results depend critically on the chosen collateral harm perspective.

Keywords: Security, Airspace Management, UAS Communications
Abstract: Secure Unmanned Traffic Management (UTM) is critical as Unmanned Aerial Vehicles (UAVs) integrate into shared airspace; perimeter-based security is insufficient for UTM’s decentralized, multi-stakeholder setting. We implement and evaluate a Zero Trust UTM framework [19] that applies National Institute of Standards and Technology (NIST) Zero Trust Architecture (ZTA) to UAV, ground station, and UTM interactions. A four-component testbed is used: QGroundControl (QGC) with UTM Adapter, Flight Blender with Policy Enforcement Point (PEP) and Policy Decision Point (PDP) and Zero Trust layers, a custom quadrotor platform, and a Technology Innovation Institute (TII) Security Verifier. We address four problems: unauthorized access (P1), Man-in-the-Middle(MitM) on links (P2), lack of per-request authorization and audit (P3), and integrity and non-repudiation (P4). Evaluation shows that with Zero Trust enabled, all tested attack types (invalid token, wrong scope, MitM fake response, MitM modify payload, replay) are blocked and logged; MitM-injected flight plan responses are rejected by signature verification using the Elliptic Curve Digital Signature Algorithm (ECDSA). Latency overhead is 25–40% versus baseline, with success rate above 99% under peak load. The framework is implementable and effective in a concrete test environment with full auditability.

Keywords: Regulations, Airspace Management, Airspace Control
Abstract: EASA's Acceptable Means of Compliance and Guidance Material (AMC/GM) to Regulation (EU) 2021/664 on a regulatory framework for the U-space put into writing the obvious fact (obvious to U-space stakeholders) that U-space airspaces must be regularly assessed based on a well worked out set of performance metrics. This paper reports on a probabilistic analysis of a variety of Key Performance Indicators (KPIs) for UAV (Unmanned Aerial Vehicle) Traffic Management (UTM). The KPIs belong to the Key Performance Areas (KPAs) of Access and Equity, Capacity, Efficiency, and Safety. The metrics are evaluated on a variety of simulated demand scenarios in very low level (VLL) metropolitan airspace over European cities. In particular, we study the dependence of the UTM KPIs on the lookahead time of strategic deconfliction of U-plans; the length of this time interval, between the deconfliction and the desired takeoff time of the drone operation, is called Reasonable Time to Act (RTTA) in the European Concept of Operations for U-space (CORUS) and its Urban Air Mobility (UAM) extension (CORUS-XUAM). In view of the stochastic nature of VLL UAV traffic, we present not only the average values of the KPIs, but also their probability distributions.

Keywords: Regulations, Swarms, Autonomy
Abstract: Autonomous drone swarms offer significant potential for civil applications such as search and rescue, disaster response, and infrastructure monitoring, yet they challenge regulatory frameworks originally designed for single unmanned aircraft under direct human control. This paper analyzes key regulatory and safety issues arising from civil autonomous drone swarms, focusing on accountability, autonomy governance, airspace integration, and public safety and privacy. Using European Union law as the primary reference, complemented by a Swedish national case study and comparative insights from selected non-EU jurisdictions, the paper identifies structural gaps in aviation, AI, and data protection regulation. It argues that compliance-by-design—embedding legal and safety requirements directly into swarm architectures and operational concepts—is essential for enabling safe, lawful, and publicly acceptable deployment of autonomous drone swarms in civil airspace.

Keywords: Training, UAS Applications, Path Planning
Abstract: In the literature, actor-critic model predictive control (AC-MPC) integrates MPC with reinforcement learning to enable high-performance control of complex dynamical systems. However, its differentiable MPC layer requires repeatedly solving an optimization problem in both the forward and backward passes, leading to substantial training and inference latency. This paper tackles this bottleneck introducing a CUDA-accelerated variant that significantly reduces end-to-end execution time while preserving the control performance of the baseline formulation. Simulation results on an agile drone racing task show that our approach achieves state-of-the-art lap times and near-limit dynamic behaviour with markedly reduced training and inference time.

Keywords: Regulations, Certification, Airworthiness
Abstract: The paper conducts a multi-dimensional analysis of licensing systems for unmanned aircraft (drone) operators in selected Member States of the European Union. The research includes a legal analysis of national regulations implementing the EU regulatory framework, in particular Regulations (EU) 2019/947 and 2019/945, as well as an assessment of the practical functioning of these systems, with particular emphasis on the degree of their harmonization with EU regulations and their impact on operational safety and the development of the drone services market. The study also seeks to identify similarities and differences between the solutions adopted in individual Member States and to assess their impact on the development of the drone services market, the level of operational safety, and the mobility of operators within the EU internal market.

Keywords: UAS Applications, Networked Swarms
Abstract: In this paper, a novel cooperative encirclement guidance method is proposed for multiple missiles against the maneuvering target. By covering the target’s reachable domain with the missiles’ combined detection range, the ideal missile number and dynamic encirclement angles are designed to improve the capture probability and block escape directions of the target. Moreover, the desired positions of missiles are calculated in real-time within the leader-follower framework. Then, considering the non-adjustability of speed, a cooperative guidance law is constructed without chattering based on the continuous finite-time stabilizing control scheme to eliminate tracking errors. The finite-time stability of the system is proved theoretically. Finally, the effectiveness and feasibility are verified by numerical simulation under the constraint of balanced maneuverability.

Keywords: UAS Applications, Swarms, Path Planning
Abstract: This paper introduces a novel Model Predictive Control (MPC) framework for multi-UAV Search and Track in maritime environments. Unlike existing approaches that treat detection and estimation as separate processes, our method embeds probabilistic target detection models and estimation uncertainty directly within the MPC formulation. This enables adaptive control decisions based on detection confidence and estimation quality. The framework addresses cooperative guidance of a UAV swarm for simultaneous search and tracking of multiple moving targets with altitude-dependent sensing, where detection probability varies with flight height. By integrating target detection probabilities as decision variables, the MPC controller generates trajectories that balance search coverage and tracking accuracy while accounting for environmental uncertainties. Quantitative simulations demonstrate improved target acquisition rates and tracking precision under maritime conditions. Real-world prototype validation confirms both computational feasibility and practical effectiveness, establishing the transition from theoretical framework to operational autonomous UAV systems for maritime surveillance.

Keywords: UAS Applications
Abstract: Aerial surveillance using Unmanned Aerial Vehicles (UAVs) faces challenges in detecting small targets, particularly under varying illumination conditions. Single-modality detectors degrade in performance when targets are visually camouflaged or when lighting is poor. To improve robustness across day/night and reduced-visibility scenarios, this paper introduces multimodal RGB–thermal perception, leveraging the complementary information provided by visible-spectrum and thermal infrared imagery. The work introduces an enhanced multimodal architecture based on DEYOLO (Dual-Feature-Enhancement YOLO), specifically optimized for small object detection in high-altitude imagery. The model employs dual RGB and thermal backbones and integrates lightweight architectural refinements aimed at improving multi-scale representation and cross-modal alignment. In particular, the feature pyramid is extended toward higher-resolution levels to better capture small targets, with SPDConv introduced as an additional module in the backbone. The neck is modified with SPANet to enhance the capture of fine-grained details, while attention-based fusion modules adaptively weight spatial and channel information across modalities. The approach is evaluated on public RGB–thermal UAV datasets and a custom dual-sensor aerial dataset. Ablation tests validate the effectiveness of these enhancements. Experimental results show that the proposed multimodal configuration improves detection accuracy and robustness compared to single-modality baselines while maintaining computational efficiency compatible with near-real-time deployment.

Keywords: UAS Applications, Perception and Cognition, Smart Sensors
Abstract: The integration of Unmanned Aerial Vehicles (UAVs) into Intelligent Transportation Systems (ITS) offers synoptic visibility for traffic monitoring, yet scalable deployment is hindered by trajectory fragmentation, where vehicle identity persistence is lost across multi-UAV Fields of View (FOV). While state-of-the-art frameworks excel in optimizing local trajectory extraction and stability for single-drone imagery, they often function as isolated data silos that generate disjointed trajectories, thereby precluding network-level analysis such as Origin-Destination estimation. This paper presents a real-time Multi-Camera Multi-Vehicle Tracking (MCMT) system designed to handle global identity persistence. Addressing the visual ambiguity and computational cost of appearance-based Re-Identification (Re-ID) in nadir views, we introduce a lightweight Topology-Based Spatiotemporal Handover mechanism. We implement a high-throughput parallel pipeline leveraging YOLO11 and ByteTrack to process concurrent 4K streams. Our core contribution is a deterministic queue-based matching algorithm that utilizes geometric overlaps and virtual lane discretization to predictively manage identity handover via FIFO queues. Experimental results on complex urban environments, including intersections and merging traffic, demonstrate a Handover Success Rate (HOSR) of 99.8% in continuous traffic flows, significantly outperforming Re-ID baselines (74.1%) while validating edge deployment feasibility. The source code is available at https://github.com/JYe9/multi-camera-multi-vehicle-tracking -system.

Keywords: Perception and Cognition, UAS Applications, See-and-avoid Systems
Abstract: Reliable detection of consumer-grade quadcopters is necessary for aerial security and collision avoidance. Deploying high-frequency detectors on edge platforms with strict Size, Weight, and Power (SWaP) budgets, however, remains difficult. This paper presents DroNet, an ultra-lightweight drone detector (0.07M parameters) inspired by the YOLOX architecture, built entirely with depthwise separable convolutions and a simplified Feature Pyramid Network backbone. DroNet achieves mAP@.5-.95 scores of 0.50--0.55 depending on input resolution and delivers inference speeds above 130 FPS on the Luxonis OAK VPU. Across resolutions (192x192 to 320x320) we observe an empirical resolution-invariance regime for the chosen sensor and target: the 192x192 variant detects 7-inch quadcopters at the same 40 m range as higher resolutions while delivering 2.1x higher throughput. Against YOLOX-Nano and YOLOX-ShuffleNetV2 at 256x256, DroNet matches long-range detection at 3–4x the speed, making it suitable for real-time drone surveillance on companion edge computers.

Keywords: Perception and Cognition, Sensor Fusion, UAS Applications
Abstract: Advancements and the widespread availability of unmanned aircraft systems (UASs) have contributed to a rise in unauthorized operations over critical infrastructure and public spaces, making the investigation of techniques for detection and tracking of rogue drones a necessity. This work presents a real-time framework for rogue drone detection, tracking, and localization by combining vision-based drone detection with laser range-finder measurements and camera gimbal state (roll, pitch, yaw). Specifically, a measurement and validation stage rejects inconsistent detections using range and camera-position geometry, while an extended Kalman filter (EKF) with a constant-velocity model fuses the validated observations to produce rogue drone tracks and to mitigate the effects of noise, outliers, and intermittent missed detections. The rogue drone estimates are then converted into geographic coordinates using the proposed system's Global Positioning System (GPS), altitude, yaw, and pitch angles, which are then exchanged with a custom-designed web-based platform for real-time map visualization and logging. A prototype system is implemented in hardware and software and extensively tested in numerous outdoor experiments, demonstrating that the proposed filtering and sensor fusion framework significantly improves detection and tracking accuracy of rogue drones inside a designated area.

Keywords: Aerial Robotic Manipulation, Control Architectures
Abstract: This work presents the first closed-loop control framework for cooperative payload transportation with non-stopping flying carriers. The proposed method includes a feedback wrench controller that actively regulates the load’s pose by computing the wrench required for tracking its desired pose trajectory. Building upon grasp-matrix formulation and internal force redundancy, an optimization layer dynamically shapes internal-force parameters to guarantee persistent carrier motion, while not altering the desired load wrench. The desired non-stopping carrier's trajectories are computed using the system's kinematics and desired cable forces. Numerical simulations demonstrate that the method successfully prevents the carriers from stopping, while achieving a successful tracking of the desired load trajectory.

Keywords: Aerial Robotic Manipulation, Multirotor Design and Control
Abstract: Omnidirectional multirotor aerial vehicles enable independent force and moment generation in arbitrary orientations. This capability makes them particularly attractive for aerial physical interaction tasks, as they can exert forces in any direction. However, the maximum achievable force typically depends on the direction, resulting in anisotropic force capabilities and preferred interaction directions. This paper presents design insights for single-axis tilting-rotor omnidirectional multirotors aimed at generating isotropic force sets, i.e., configurations in which the maximum achievable force is independent of direction, thereby eliminating preferred interaction axes. To this end, we introduce the Isotropic Force Space Index (IFSI) to quantify the degree of force isotropy of a platform. To avoid solving a high-dimensional optimization problem, the motor placement problem is analyzed as a composition of individual motor thrust sets. Based on this geometric interpretation, we propose the use of the Fibonacci lattice for motor placement, ensuring that the motors are evenly distributed around the platform. The proposed methodology is validated for different numbers of motors, demonstrating that the resulting platforms achieve higher force isotropy compared to state-of-the-art omnidirectional designs.

Keywords: Aerial Robotic Manipulation, Multirotor Design and Control
Abstract: Thrust saturations limit the wrench that multirotor aerial vehicles can develop. Geometric control on SE(3) can be adapted to account for actuator saturations by prioritizing the position tracking with respect to attitude control. When the pose tracking requires an unfeasible force to be generated by the actuators, the proposed controller modifies the orientation of the multirotor to preserve position tracking. This controller is also compatible with reconfigurable and morphing multirotors that can switch smoothly between under-actuation and full actuation. Simulations and experiments validate the proposed approach for multiple platforms.

Keywords: Aerial Robotic Manipulation, Reliability of UAS, Multirotor Design and Control
Abstract: This work addresses the problem of fault tolerance in aerial manipulators, focusing on maintaining maneuverability after the failure of a single rotor without adding any extra hardware. Instead, the proposed approach exploits the ability to reposition the onboard manipulator or payload, thereby modifying the vehicle’s center of mass (CoM) and moments of inertia to recover controllability. This study develops a Gain Scheduled (GS) adaptive controller that explicitly adapts to variations in the vehicle’s moment of inertia induced by payload repositioning. The controller is implemented on a resource-constrained autopilot, which requires a design that balances computational efficiency with robustness. Simulations and experimental flight tests validate the proposed method, comparing the Gain Scheduled controller to a conventional PID controller under similar maneuvering scenarios.

Keywords: Aerial Robotic Manipulation, Biologically Inspired UAS, UAS Applications
Abstract: Continuum manipulators mounted on multi-rotor UAVs enable compliant aerial manipulation, but payloads and propeller downwash amplify out-of-plane bending and twisting that degrade end-effector pose accuracy. To address this problem, we present a universal-joint-based continuum manipulator designed to improve resistance to out-of-plane deformation during aerial manipulation. The proposed design uses a tubular backbone with spring-reinforced universal joints and an integrated conduit for internal routing and fluid delivery. We evaluate the design in still air and under peak propeller downwash across varying payloads, and benchmark it against a prior Nitinol-backbone CM. Bench tests show improved resistance to out-of-plane deformation across all conditions. Under peak downwash, the proposed design reduces mean error by 2.5–4× in yaw, 2–45× in y-axis, and up to 5× in roll compared to the NiTi-backbone design. We further analyze hover stability through in-flight coupled-disturbance tests over varying payloads and actuation speeds, and demonstrate the system in water sampling, spot spraying, and object transport.

Keywords: Sensor Fusion, Perception and Cognition
Abstract: Accurate infrastructure-based UAV localization remains challenging in the presence of occlusions, clutter, and limited observability from single sensing modalities. We present a distributed multi-station tracking framework that fuses mmWave radar and monocular vision to achieve robust 3D position and velocity estimation. Building upon a prior single-station radar–vision tracker, we extend the approach to a network of three portable observation stations, each performing local multi-hypothesis tracking with uncertainty-aware Kalman filtering. Vision measurements provide angular constraints that improve radar data association and mitigate clutter-induced artifacts. Instead of transmitting raw detections, each station communicates a compact state estimate and covariance to a central fusion node, where an information-form filter produces a globally consistent estimate. The system is validated in indoor flight experiments using motion capture ground truth, while remaining fully indepen- dent of it during estimation. The fused solution achieves a 3D RMSE of 0.2342 m and improves robustness against degraded individual station estimates. These results highlight the potential of distributed radar–vision sensor networks for scalable and reliable infrastructure-based UAV localization.

Keywords: Fail-Safe Systems, Navigation, Levels of Safety
Abstract: One of the most crucial functionalities of a UAV is to always be aware of its location, which is usually provided by an established navigation system, such as GNSS. However, there is no fallback mechanism in the case where GNSS is denied. For this reason, we propose a framework that acts as a fallback system for such occasions and relies mainly on the visual input from the UAV's camera. At its core, our framework uses this visual input and tries to match the drone's current POV with the correct area of a pre-loaded satellite map. Instead of using raw images, our approach relies on high-dimensional image embedding vectors generated using transformer neural networks and, more specifically, DINOv2. For matching, we utilize a codebook trained on samples from the satellite map, where we aggregate local image features using the VLAD algorithm and create image descriptors. Localization is then achieved by comparing the similarity between the UAV's live visual descriptor and the stored descriptors of the map. Our framework yielded promising results, and our analysis indicates that it is lightweight enough to operate on a UAV.

Keywords: Sensor Fusion, Perception and Cognition, Autonomy
Abstract: Indoor autonomous drones are increasingly adopted for inspection, monitoring, and inventory tasks in GNSS-denied facilities, where reliable global positioning is essential to automate repeatable takeoff, navigation, and landing. A key challenge is enabling accurate relocalization to a global reference frame when the docking station can be placed arbitrarily, without relying on visual markers or infrastructure-intensive site modifications. To address this problem, we present a global localization system for indoor Unmanned Aerial Vehicles (UAVs) that enables takeoff and landing operations from a portable docking station ('drone-in-a-box'). Our approach fuses Ultra-Wideband (UWB) ranging and Light Detection and Ranging (LiDAR) measurements within an Error-State Extended Kalman Filter (ES-EKF), and is sensor-model agnostic, supporting Inertial Measurement Unit (IMU), LiDAR, and different hardware configurations. A UWB tag mounted on the drone interacts with anchors installed in the environment and additional tags on the docking box to provide robust range-based measurements; these enable an initial pose estimate computed via Particle Swarm Optimization (PSO), which is subsequently refined through LiDAR scan matching to refine localization. Experimental results in real-world indoor scenarios validate the proposed system, demonstrating accurate and repeatable autonomous deployment and recovery in indoor environments.

Keywords: Perception and Cognition, Navigation
Abstract: Visual place recognition (VPR) can be used for aerial geo-localization (VG), by matching aerial images against a database of geo-referenced satellite images. VG can be a valuable replacement or complement for navigation within environments where the use of Global Navigation Satellite Systems (GNSS) is denied. While deep learning methods have significantly advanced VPR performance, most provide point estimates without statistical guarantees. We apply the Conformal Prediction (CP) framework to retrieval-based VPR, formulating distance-based and cumulative similarity-based nonconformity measures that generate retrieval sets guaranteed to contain the true location with a user-specified probability. While prior conformal retrieval work requires uncertainty-aware feature extractors, limiting applicability to specific model families, we propose model-agnostic formulations that operate directly on metric space, enabling the use of any pretrained backbone. We further propose a novel uncertainty measure for fixed-size retrieval sets, based on CP, enabling uncertainty-aware decision making under fixed computational budgets. Experiments on two cross-view aerial/satellite datasets validate coverage guarantees and compare the efficiency of standard CP, Adaptive Prediction Sets, and calibrated fixed-k retrieval.

Keywords: Sensor Fusion, Perception and Cognition, Navigation
Abstract: This paper investigates the feasibility of using an onboard ultrasonic anemometer for ground velocity estimation on a multirotor Unmanned Aircraft System (UAS), framed as a proof-of-concept study. The sensor is considered as an additional sensing modality that may complement conventional navigation sources, particularly in Global Navigation Satellite System (GNSS)-degraded environments. While anemometers directly measure airspeed rather than ground speed and are affected by rotor-induced airflow disturbances, they offer attractive properties such as immunity to lighting conditions and resistance to external radio interference. A data-driven approach is proposed to compensate for aerodynamic effects and wind influence. A compact multilayer perceptron with a sliding temporal window is trained to map measured signals to reference ground velocity. The study includes the design and integration of a dedicated hardware setup and the collection of a real-flight dataset comprising numerous data collection missions and almost one hundred thousand samples. Experimental evaluation demonstrates a consistent reduction of bias and root mean square error compared to raw airspeed measurements, confirming the viability of the proposed concept. The estimator remains computationally lightweight, with very low inference time on standard CPU hardware, supporting real-time onboard deployment.

Keywords: Perception and Cognition, Sensor Fusion
Abstract: Event cameras provide high-temporal-resolution visual sensing that is well suited for observing fast-moving aerial objects; however, their use for drone trajectory prediction remains limited. This work introduces an event-only drone forecasting method that exploits propeller-induced motion cues. Propeller rotational speed is extracted directly from raw event data and fused within an RPM-aware Kalman filtering framework. Evaluations on the FRED dataset show that the proposed method outperforms learning-based approaches and vanilla Kalman filter in terms of average distance error and final distance error at 0.4s and 0.8s forecasting horizons. The results demonstrate robust and accurate short- and medium-horizon trajectory forecasting without reliance on RGB imagery or training data.

Keywords: Perception and Cognition, Autonomy, Sensor Fusion
Abstract: Reliable monocular ranging is a critical capability for autonomous probe-and-drogue aerial refueling under size, weight, power, and certification constraints. Reported performance across monocular ranging approaches is often difficult to compare due to differences in datasets, calibration, ground truth, and metrics. This paper presents a fair benchmark of two geometric methods: (i) a DNN detection pipeline that estimates 3D relative position from bounding-box geometry using calibrated similar-triangle relationships with weighted fusion, and (ii) a segmentation-driven pipeline that estimates 3D relative position by fitting an ellipse to the drogue rim and applying pinhole projection. Both approaches are evaluated on the same dataset with identical calibration, shared motion-capture ground truth, and unified error definitions. Performance is assessed using signed-error box plots and RMSE curves across a 7–22 ft envelope, along with normalized error distributions to support comparison across distance. Results show complementary behavior: the bounding-box method exhibits range-dependent bias drift, increasing RMSE with distance, and more frequent long-range outliers, whereas the segmentation/ellipse method maintains more uniform mid-range performance with a consistent overestimation bias but shows an abrupt far-field degradation where rim pixel support is reduced. These findings clarify accuracy–robustness tradeoffs and guide method selection for embedded AAR systems.

Keywords: Perception and Cognition, Autonomy, Sensor Fusion
Abstract: Autonomous aerial refueling (AAR) can substantially extend the endurance and operational reach of uncrewed aerial systems (UAS), but many approach-phase concepts still rely on GPS-aided cues for join-up and deconfliction. Since GPS availability is not assured in contested environments, this paper evaluates a monocular, geometry-based alternative for relative navigation using multi-landmark pose recovery. A YOLO26 detector localizes six physically marked landmarks on a scaled aircraft surrogate, providing 2D image measurements paired with known 3D landmark coordinates. Camera-to-target pose is recovered via Perspective-textit{n}-Point (PnP) using OpenCV texttt{solvePnP}, and the reported range is the optical-axis depth extracted from the estimated translation vector. Experiments are conducted in the USNA VIPER Laboratory with motion-capture ground truth, enabling repeatable evaluation across stand-off distance and centerline/pm 15^{circ} viewing angles. Detector performance is reported using precision--recall behavior and mean average precision (mAP), while ranging accuracy is quantified using mean absolute error (MAE) and RMSE. Results show an approximately linear depth-to-truth relationship with perspective-dependent systematic bias in the raw estimates; a simple per-perspective affine calibration reduces aggregate error from 24.77~mm MAE to 7.52~mm MAE (27.85~mm to 10.01~mm RMSE) over 1--16~ft in-lab distances. These findings provide measurement-driven evidence for the viability of multi-keypoint monocular ranging and clarify dominant error modes relevant to scaling toward longer equivalent stand-off distances under controlled conditions.

Keywords: Control Architectures, Navigation, Perception and Cognition
Abstract: The United States Navy plans to increase the number of uncrewed aerial systems (UASs) within its carrier air wings. These UASs are expected to operate in GPS-denied and RF-denied environments, requiring fully autonomous capabilities. A key enabler is the ability to perform autonomous aerial refueling without reliance on external positioning systems. This research investigates the viability of using relative position information from a single camera to autonomously control an UAS for aerial refueling docking. A deep neural network (DNN)-based monocular ranging system was adapted and evaluated in the University of Maryland’s Omni-domain Autonomous Systems Integration Space (OASIS), a 4,800 square foot motion capture environment. A YOLO11m DNN was trained to identify the aerial refueling coupler and drogue basket at ranges from 5–50 ft. A similar-triangles ranging algorithm produced real-time position estimates of the coupler relative to the UAS that were utilized for navigation control and compared against truth data. The UAS used the position estimations and a bang-bang control architecture to align itself to an F/A-18 refueling drogue and approach the docking position. Across four test flights starting with 20 feet of range, the UAS successfully navigated to the coupler docking position. Mean error values between the estimated and true position of the drogue coupler relative to the UAS were less than 2.75 inches in the X, Z, and Range components, and less than 4.75 inches in the Y component. These results demonstrate that a single monocular optical payload can be used as a closed-loop control input for autonomous docking during AAR.

Keywords: Certification
Abstract: The future of transportation is uncrewed and ultimately autonomous. Yet, a clear pathway does not exist on how to get there. The 4 February 2026 Development, Test & Evaluation, Verification & Validation (DTEVV) of Autonomous Systems Workshop focused on mapping out the priority research and workforce challenges that need to be addressed before it will be possible to field truly autonomous technology. The DTEVV Workshop brought together more than 100 technology and operations experts, faculty, students, members of the DTEVV workforce, and policymakers from academia, industry, and government to help define the gaps and offer strategies that may enable the fielding of autonomous systems. The morning focused on research gaps that need to be filled to enable the fielding of an autonomous system. The afternoon focused on the needs of the DTEVV workforce as they are tasked with evaluating the risks associated with fielding these systems. Both the morning and afternoon consisted of a keynote address, an interactive panel, and a proctored road map development activity. In total, 44 road maps were developed by the workshop attendees, and they are currently being analyzed to help guide future decisions on pathways for fielding autonomous systems.

Keywords: Autonomy, Certification, Interoperability
Abstract: The integration of artificial intelligence agents into operational aerospace platforms presents a set of testing, trust, and interoperability challenges that existing certification and validation frameworks were not designed to address. This paper examines the X-62 VISTA (Variable Inflight Simulator Test Aircraft) as a case study in transitioning AI from laboratory demonstration to flight-ready operation. Rather than proposing definitive solutions, this work frames three interrelated research questions that demand attention as AI technologies mature toward fielded systems: (1) how to define and validate acceptable operational envelopes for nondeterministic AI systems, (2) how human operators should calibrate trust in agents that process information beyond human perceptual bandwidth, and (3) how networked intelligent systems will behave when operating within shared, contested airspace. Drawing on recent X-62 program milestones and the broader landscape of autonomous systems research, the paper argues that the X-62 VISTA offers a uniquely suitable platform for exploring these questions within a safe, bounded, and operationally relevant flight-test environment.

Keywords: UAS Applications, Perception and Cognition
Abstract: The lack of reliable annotated datasets remains a major obstacle to applying deep learning in agriculture. This issue is particularly pronounced for large-scale orthomosaics, where manual labeling is impractical. In many agricultural settings, initial annotations are derived from available domain knowledge, such as planned crop layouts. While this information provides useful structural guidance, it often results in labels with substantial noise when actual plant emergence deviates from expectations. In this work, we investigate whether existing label-noise handling techniques can be adapted to generate usable training labels under such conditions. We propose a practical annotation pipeline for agricultural orthomosaics that combines prior planting information, unsupervised clustering, and iterative refinement using Co-teaching and Co-teaching+. The resulting Iterative Co-teaching+ pipeline progressively reduces the proportion of incorrect labels through domain-informed corrections and iterative training. On a newly introduced Oilseed Radish and Winter Rapeseed (OR--WR) dataset, initial labels with an estimated noise level exceeding 80% are refined to below 6%. Rather than proposing a new learning algorithm, this work examines the limits of existing methods under extreme label noise and demonstrates how domain knowledge can be used to stabilize label refinement in practice. To assess the practical impact of label quality, we train commonly used classifiers (YOLOv11n and ResNet-50) using labels obtained at different stages of the pipeline. This enables an analysis of how residual label noise affects downstream performance and the extent to which it can be tolerated. In addition to the proposed pipeline, we release the OR--WR dataset and a QGIS plugin to support efficien

Keywords: UAS Applications, Control Architectures
Abstract: This paper presents a non-contact approach for vibration-based structural damage detection using an autonomous and customized cost-effective unmanned aerial vehicle (UAV). The developed UAV integrates a lightweight onboard camera, embedded processing, and a visual positioning algorithm to enable stable operation and target tracking in indoor environments. Vibration signals are extracted from video recordings through vision-based motion tracking to identify shifts in natural frequencies indicative of structural degradation. A laboratory-scale frame structure is evaluated under healthy and simulated-damage conditions, where damage is introduced via an added mass. The proposed system is validated through a multi-platform experimental study involving two high-resolution smartphones, a USB camera, and a custom-built low-cost UAV equipped with an onboard camera and an AprilTag-based autonomous alignment system for operation in GPS-denied environments. The displacement time is extracted and analyzed in the frequency domain and compared to reference measurements from contact accelerometers and a finite element model. Experimental results from two contemporary mobile devices, a USB camera, and a custom-developed UAV show that all platforms successfully capture the fundamental frequency and its shift due to damage. Although the UAV exhibits slightly higher errors (approximately 5–6%) due to platform-induced disturbances and sensing limitations, it reliably detects damage-induced frequency changes. Compared to commercial UAV systems, the proposed platform achieves comparable inspection performance at significantly lower cost. These results demonstrate that low-cost autonomous UAVs provide a practical, flexible, and scalable solution for structural health monitoring.

Keywords: UAS Applications, Payloads, Perception and Cognition
Abstract: Time-critical response missions require rapid geolocated target reports and scene context to support fast decisions. Unmanned aerial vehicles (UAVs) are well suited for wide-area scanning and target search, but aerial video can be difficult to interpret reliably under clutter, occlusions, and time pressure. To address this, we present USPI, a modular ROS~2 UAV Sensor Payload Interface that supports operator-in-the-loop use of live RGB/thermal streams. USPI enables click-to-center gimbal pointing and a pause-and-annotate workflow to localize one or more targets from image clicks, and it maintains a map-aligned mosaic overlay that updates scene context in real time. We evaluate USPI in controlled 15~m AGL field tests and in a timed mock mass-casualty scenario at 30~m AGL, showing sub-meter localization accuracy in the controlled tests and a median error of 1.3~m (RMSE 1.7~m) in the mock scenario.

Keywords: UAS Applications, Autonomy, Aerial Robotic Manipulation
Abstract: This work presents an integrated control and software architecture that enables arguably the first fully autonomous, contact-based non-destructive testing (NDT) using a commercial multirotor originally restricted to remotely-piloted operations. To allow autonomous operation with an off-the-shelf platform, we developed a real-time framework that interfaces directly with its onboard sensor suite. The architecture features a multi-rate control scheme: low-level control is executed at 200 Hz, force estimation at 100 Hz, while an admittance filter and trajectory planner operate at 50 Hz, ultimately supplying acceleration and yaw rate commands to the internal flight controller. We validate the system through physical experiments on a Flyability Elios 3 quadrotor equipped with an ultrasound payload. Relying exclusively on onboard sensing, the vehicle successfully performs autonomous NDT measurements within an unstructured, industrial-like environment. This work demonstrates the viability of retrofitting off-the-shelf platforms for autonomous physical interaction, paving the way for safe, contact-based inspection of hazardous and confined infrastructure.

Keywords: Path Planning, UAS Applications, Control Architectures
Abstract: This paper introduces inspection through GLASS, a Geometric Look-Angle Shaping Strategy for enclosed regions using unmanned aerial vehicles. In doing so, the vehicle’s guidance command is constructed through a bounded, look-angle shaping function relative to a desired standoff path. By embedding a smooth, hyperbolic-tangent-type shaping function within a polar geometric framework, GLASS ensures global existence of the guidance dynamics while avoiding the far-field limitations inherent to conventional formulations. Lyapunov stability analysis establishes asymptotic convergence to a prescribed inspection standoff under explicit curvature feasibility conditions, along with analytical settling-time characteristics. The proposed strategy incorporates maximum turn-rate constraints without inducing singularities throughout the workspace. High-fidelity six-degree-of-freedom quadrotor simulations demonstrate the effectiveness of GLASS in representative enclosed inspection scenarios, highlighting a practically viable guidance framework for autonomous enclosed inspection missions.

Keywords: UAS Applications, Perception and Cognition
Abstract: In this paper, we conduct the first work to revolutionize weed detection in residential lawns using unmanned aerial vehicles (UAVs). Compared with prior weed detection studies, our target scenario is much harder. Weeds become extremely small (from near-ground cameras to UAVs) and visually similar to surrounding turfgrass (from crop fields to residential lawns), making them difficult to distinguish. Moreover, the lack of annotated aerial lawn weed datasets forces detectors to be trained on public ground-level datasets, introducing a ground-to-aerial domain shift that significantly degrades detection accuracy. To address these challenges, we propose A4AWD, a lightweight inference-time augmentation tailored for aerial weed detection. A4AWD devises a new weed-specific augmentation method tailored for green-on-green detection, where weeds are extremely tiny, irregular, and visually similar to the surrounding turfgrass. Our field test over 165 residential houses shows that A4AWD outperforms three SOTA approaches and achieves high precision in both healthy and weedy lawns, roughly 1.7x-2.5x higher than the second best. Our dataset is released at Github [1].

Keywords: Aerial Robotic Manipulation, Multirotor Design and Control, UAS Applications
Abstract: This paper presents a novel framework for au- tonomous inspection of metallic infrastructure leveraging real- time stiffness estimation, surface-aligned impedance adaptation in underactuated aerial unmanned aerial vheicle (UAV) and unactuated manipulatior, addressing the pervasive challenge of underactuation-induced sliding during contact tasks. Sim- ulation results in Gazebo demonstrate suppression of lateral sliding and precise force regulation on inclined planes, enabled by real-time stiffness adaptation through a Lyapunov-derived impedance law. Field trials confirm robust contact maintenance, accurate adaptive stiffness convergence, and reliable inspection under variable surface geometries and wind disturbances. A LiDAR-based surface detection and RANSAC pipeline aligns the UAV control frame with the local surface normal, de- coupling normal-force regulation from underactuated lateral dynamics. Together with a compliant end-effector integrating a load cell and passive adhesion, this design cushions impacts while preserving accurate force sensing. The integration of surface-aligned impedance adaptation, online stiffness estima- tion, and lightweight mechanical compliance yields an unmod- ified, cost-effective underactuated UAV platform capable of autonomous and stable surface interaction, paving the way for rapid, safe, and high-quality industrial inspection workflows.

Keywords: Aerial Robotic Manipulation, Perception and Cognition, Multirotor Design and Control
Abstract: This paper investigates two distinct approaches to the dynamic modeling of aerial continuum manipulators (ACMs): the decoupled and coupled formulations. Both open-loop and closed-loop behaviors of a representative ACM are analyzed. The primary objective is to determine the conditions under which the decoupled model attains accuracy comparable to the coupled model while offering reduced computational cost under identical numerical conditions. The system dynamics are first formulated using the Euler–Lagrange method under the piecewise constant curvature (PCC) assumption, with explicit treatment of the near-zero curvature singularity. A decoupled model is then obtained by neglecting the coupling terms in the ACM dynamics, enabling systematic evaluation of open-loop responses under diverse actuation profiles and external wrenches. To extend the analysis to closed-loop performance, a novel dynamics-based proportional-derivative sliding mode image-based visual servoing (DPD-SM-IBVS) controller is developed to regulate image feature errors in the presence of a moving target. The controller is implemented with both the decoupled and coupled models, allowing a direct comparison of their effectiveness. Open-loop simulations reveal pronounced discrepancies between the two modeling approaches, particularly under varying torque inputs and continuum arm parameters. Conversely, closed-loop experiments demonstrate that the decoupled model achieves tracking accuracy on par with the coupled model (within subpixel error) while incurring lower computational cost.

Keywords: Aerial Robotic Manipulation, UAS Applications
Abstract: Soft aerial manipulation is often hindered by structural gripper mass and payload-induced sway, which makes UAV flight control difficult. This paper introduces a monolithic, tendon-driven soft gripper utilizing Functionally Graded Lattices (FGL) 3D-printed from TPU. By modulating unit-cell geometry in fingers, we achieve localized mechanical properties programming while eliminating the pneumatic complexity of state-of-the-art alternatives. A systematic parametric study on a single soft finger identified a 10 mm hexagonal lattice finger as the optimal architecture, yielding a 75% mass reduction and an approximately 100% improvement in vibration damping (1.0s vs. 2.0s settling time) compared to silicone benchmarks. Integrated via a modular mechatronic system, the 150 g soft gripper assembly successfully transported payloads up to 650 g. Our results demonstrate that architected metamaterials provide a robust, lightweight, and rapidly manufacturable pathway for stable, high-speed soft aerial grasping in complex UAV applications.

Keywords: Aerial Robotic Manipulation, Biologically Inspired UAS, Energy Efficient UAS
Abstract: Micro Aerial Vehicles (MAVs) have advanced considerably in recent years due to their versatility and ability to operate in cluttered environments. However, their real-world use remains limited by endurance constraints and the difficulty of carrying large batteries. Existing solutions include perching, which often rely on motors, sensors, or complex actuation, adding weight, consuming energy, being surface-specific, and limiting integration on small MAVs. In contrast, this study presents a lightweight, fully passive perching mechanism that enables both planar and cylindrical landings without energy consumption, instead leveraging the vehicle’s own weight for secure, reliable engagement. The gripper uses a rack-and-pinion mechanism for simplified actuation, and it disengages once the weight is lifted. This design ensures stable perching with minimal energy loss during perching and takeoff. With a single gripper weighing just 33 g, capable of being integrated on vehicles more than 20 times its weight, the mechanism provides perching on both cylindrical and planar surfaces, thus significantly enhancing the current capabilities of aerial robots. Testing was done on three different platforms: a 1.3 kg DJI drone, a 102 g entomopter, and a 450 g ornithopter. Tests showed the adaptability of the mechanism to different vehicles and its effectiveness in increasing the overall mission duration in each case.

Keywords: Aerial Robotic Manipulation, Autonomy, Navigation
Abstract: This paper presents a decentralized guiding vector field framework for rigid formation path following in multi-UAV systems based on the superposition of simple rule-driven vector fields. Each control objective is encoded as a dedicated vector-field component: a normalized tangential term to ensure progression along a prescribed geometric path at a desired speed, a convergence term driving the formation toward the path, a local coordination term that preserves the desired formation shape through a pseudo-path parameter, and a local separation term for inter-agent collision avoidance. The coordination mechanism requires the exchange of only a single scalar per agent, resulting in low communication bandwidth. Partial theoretical results establish convergence of the swarm to the desired path and formation in the absence of the separation term, while collision avoidance is formally guaranteed. The approach is experimentally validated using a swarm of five quadrotors executing two 3D choreographies. The first experiment involves tracking a helical path while maintaining a rigid V-shaped formation. The second demonstrates a spiral-like motion obtained by a rotating circular formation tracking a vertical line. The experimental results support the effectiveness and practical applicability of the proposed framework.

Keywords: Simulation, UAS Testbeds, Perception and Cognition
Abstract: Developing and testing UAV algorithms for Antarctic operations is constrained by brief summer field windows, sub-zero equipment failures, and the high cost of remote deployments. We integrate real-world terrain data from three East Antarctic sites (ASPA~135, Robinson Ridge, and Bunger Hills) into Gazebo simulation environments with accurate terrain geometry, RGB textures, and thermal signatures derived from field sensor data. The pipeline introduces a thermal terrain modelling method that compensates for Gazebo's lack of thermal heightmap support by generating transparent mesh overlays from digital elevation models with embedded thermal properties, enabling thermal search and rescue simulations over varied Antarctic terrain. The pipeline processes field-collected imagery through automated orthomosaicking, raster alignment, and texture generation stages, producing simulation-ready assets at ground sampling distances from 0.25 to 10.8~cm/pixel. A thermal search and rescue case study validates the pipeline: a YOLOv8 detector trained entirely on simulated thermal imagery achieved 0.995 mAP@0.5 and 1.0 recall for human target detection across all three terrains. The framework integrates ROS2 Humble, PX4 Software-in-the-Loop autopilot, and real-time object detection, supporting applications including autonomous navigation and environmental monitoring. All simulation environments and tools are designed for reproducible Antarctic UAV research, reducing reliance on costly and logistically constrained field campaigns.

Keywords: Simulation, Path Planning, Airworthiness
Abstract: The use of Unmanned Aerial Vehicles (UAVs) has grown substantially in recent years, owing to their efficiency and versatility. These advantages have led to their widespread adoption in applications such as logistics, infrastructure inspection, and environmental monitoring. However, such operations entail a high level of risk, as UAVs are particularly sensitive to adverse environmental conditions, especially wind gusts and turbulence, which can compromise flight stability and safety. This issue is especially relevant in urban areas, where the interaction between wind and built structures generates complex flow patterns, including instabilities and wake regions. These phenomena occur at spatial scales of only a few metres, requiring high-resolution tools such as Computational Fluid Dynamics (CFD) models to anticipate their formation. Consequently, recent studies have integrated CFD simulations into UAV path-planning tools to avoid flight through highly turbulent regions. Typically, these approaches analyse fluid variables such as wind speed or turbulent kinetic energy and restrict UAV access in areas exceeding predefined turbulence thresholds. However, such threshold-based methods are often arbitrary and do not account for the specific characteristics of the UAV when assessing the impact of turbulence on controllability. In this work, a turbulence assessment model is proposed to characterise UAV controllability under varying wind conditions. The proposed approach employs CFD simulation results and conducts a statistical analysis based on Dryden turbulence models. This information is then integrated into an efficient path-planning framework, enabling the computation of safe UAV trajectories in complex urban environments.

Keywords: Simulation
Abstract: One of the main limitations of multirotor UAVs is their short flight time due to battery constraints. A practical solution for continuous operation is to power the drone from the ground via a tether. While this approach has been demonstrated for stationary systems, scenarios with a fast-moving base vehicle or strong wind conditions require modeling the tether forces, including aerodynamic effects. In this work, we propose two complementary approaches for low-latency quasi-static tether modeling with aerodynamics. The first is an analytical method based on catenary theory with a uniform drag assumption, achieving very fast solve times below 1~ms. The second is a numerical method that discretizes the tether into segments and lumped masses, solving the equilibrium equations using CasADi and IPOPT. By leveraging initialization strategies, such as warm starting and analytical initialization, low-latency performance was achieved with a solve time of 5~ms, while allowing for flexible force formulations. Both approaches were validated in real-world tests using a load cell to measure the tether force. The results show that the analytical method provides sufficient accuracy for most tethered UAV applications with minimal computational cost, while the numerical method offers higher flexibility and physical accuracy when required. These approaches form a lightweight and extensible framework for low-latency tether simulation, applicable to both offline optimization and online tasks such as simulation, control, and trajectory planning.

Keywords: Simulation, Aerial Robotic Manipulation, Training
Abstract: This tutorial paper explains how to develop a ROS2 Gazebo simulator of a dual arm aerial manipulation robot intended to conduct parcel delivery operations in a representative intra-logistics scenario. This work follows a real-to-sim approach in the sense that the simulator replicates the aerial robot, scenario, and operation carried out by a fully-autonomous dual arm aerial delivery robot cite{suarez2025fully} during the euROBIN Project Cooperative Competition. The aerial robot consists of a quad-rotor platform controlled with the PX4 flight control software, equipped with a lightweight and human-size dual arm manipulator providing two joints per arm (shoulder and elbow pitch flexion/extension), integrating a camera for parcel detection and localization with Aruco markers, and a 2D LiDAR for localization and mapping. The paper outlines the process of creating the ROS2 simulator package from the 3D model and physical parameters of the robot and the objects in the scenario relying on the examples from PX4, including a 3D mesh of the flying arena scanned from the real scenario. The purpose of this work is to serve as guideline for students and young researchers in the development of an aerial robot simulator.

Keywords: UAS Testbeds, Control Architectures, Multirotor Design and Control
Abstract: The Soft-PVTOL is a planar vertical take-off and landing aircraft with soft continuum arms actuated by tendons. In its nominal configuration, the system has four control inputs two thrusts and two tendon torques and five degrees of freedom. This paper investigates the behavior of the Soft-PVTOL when the tendon actuation is lost, as may occur due to tendon failure, hysteresis, fixed tendons, or mechanical faults. In this degraded mode, the only available control inputs are the two thrust forces at the arm tips, while the arm curvatures evolve passively under the combined effect of elastic restoring forces, damping, gravity, and a thrust-induced bending torque T_i cdot l_i cdot sinc(q_i). We show that this passive-arm configuration remains locally controllable at hover for all positive values of the arm stiffness, despite losing two of its four actuators. However, the degree of controllability measured by the minimum singular value of the controllability matrix depends critically on the stiffness parameter, with a pronounced weakening near a critical threshold. Simulation results demonstrate successful takeoff and hover stabilization using a simple PD-based controller. Experimental results on a constrained test stand validate pitch tracking and passive arm dynamics under realistic operating conditions.

Keywords: UAS Testbeds, Autonomy, Simulation
Abstract: Unmanned aerial vehicle (UAV) research requires the integration of cutting-edge technology into existing autopilot frameworks. This process can be arduous, requiring extensive resources, time, and detailed knowledge of the existing system. ROSplane is a lean, open-source fixed-wing autonomy stack built by researchers for researchers. It is designed to accelerate research by providing clearly defined interfaces with an easily modifiable framework. Built around ROS 2, ROSplane allows for rapid integration of low or high-level control, path planning, or estimation algorithms. A focus on lean, easily-understood code and extensive documentation lowers the barrier to entry for researchers. Recent developments to ROSplane improve its capacity to accelerate UAV research, including the transition from ROS 1 to ROS 2, enhanced estimation and control algorithms, increased modularity, and an aerodynamic modeling pipeline. This aerodynamic modeling pipeline can reduce the effort of transitioning from simulation to real-world testing by not requiring costly system identification or computational fluid dynamics tools. ROSplane's architecture reduces the effort required to integrate new research tools and methods, expediting hardware experimentation.

Keywords: UAS Testbeds, UAS Applications, Airworthiness
Abstract: Autonomous uncrewed aerial systems (UAS) research increasingly relies on perception-driven navigation and control, yet transitioning autonomy algorithms from simulation and bench testing to repeatable flight testing remains hindered by fragile integration paths and a lack of reusable interfaces between autonomy software and flight controllers. This paper presents a flight-tested evaluation of a standardized autonomy-to-flight-controller interface implemented on a modular quadrotor research platform, extending prior work on a portable ROS~2-to-PX4 software stack. In addition, informed by a survey of comparable autonomy-capable quadrotor platforms, the paper proposes an NDAA-compliant electronics architecture that prioritizes robust offboard communication, sufficient onboard vision compute, and manageable mass/power for repeatable perception-driven experiments. An incremental indoor flight-test campaign was conducted to assess end-to-end command execution fidelity, mode-transition predictability, and operational reliability factors that emerge only in flight. Tests progressed from bench-level verification and single-setpoint offboard control to multi-setpoint trajectories and perception-driven visual navigation using fiducial marker tracking to generate closed-loop velocity commands. Results demonstrate stable and consistent tracking of autonomy-generated velocity setpoints during autonomous target approach, with reliable engagement and disengagement of offboard mode and no observed interface-related faults under nominal conditions. The paper further documents flight-specific failure cases—including low-altitude optical-flow estimation degradation, vibration sensitivity, and companion-computer communication bottlenecks—and the mitigations applied to restore reliable test.

Keywords: Autonomy, Navigation, Levels of Safety
Abstract: We propose an architecture for integrating high-level, human-provided safety rules and operator-aligned semantic preferences into autonomous robot navigation in unstructured outdoor environments. In our approach, natural-language rules are translated into Signal Temporal Logic (STL) specifications that guide planning and navigation during runtime. Persistent, environment-centric rules and terrain preferences are grounded into a 2D cost map, while temporally dynamic requirements are expressed as STL specifications to be monitored during runtime. We hypothesize the use of Vision-Language Models (VLMs) for zero-shot scene understanding, enabling mapping between human instructions, semantic features, and environmental constraints. Within this framework, we construct an illustrative navigation model that is designed to satisfy a set of STL-encoded specifications and soft operator preferences through formal satisfaction metrics embedded into environmental properties and runtime monitoring.

Keywords: Regulations, Risk Analysis, Navigation
Abstract: Uncrewed Aerial Systems (UAS) operations present a growing risk to critical infrastructure due to unexpected anomalies in navigation systems or failure of critical components. This risk is particularly elevated when operations occur in complex airspace environments near critical infrastructure. According to the Department of Homeland Security (DHS), airports and associated facilities such as air traffic control towers are classified as critical infrastructure within the transportation sector. As UAS operations become increasingly prevalent, it is essential to address the associated collision risks when critical infrastructure is present within the operational environment. Unintended UAS intrusions into restricted zones can result from flight control or subsystem failures, posing a serious threat to airport operations and human life. In this paper, we examine the collision risk of UAS operating near critical infrastructure in beyond visual line of sight (BVLOS) conditions in and around airport environments. The analysis presented is based on an air traffic control tower located at Grand Forks International Airport. Risk assessment was conducted using a digital replica of the infrastructure and surrounding airspace within a simulation environment, with risk analysis focused on loss-of-control scenarios.

Keywords: UAS Applications, Path Planning, Autonomy
Abstract: Using commercially-available, off-the-shelf equipment, a multi-agent autonomous search and rescue (SAR) system is developed to locate, identify, and maneuver to a target representing a person lost at sea. The system includes an Apache3 uncrewed surface vessel (USV) and Hexsoon EDU-650 uncrewed aerial vehicle (UAV). The UAV uses received signal strength indicator (RSSI) data to localize a radio frequency (RF) source mounted on the target and communicates the target’s GPS coordinate to the USV. A combined steering and heading control is developed for the USV to navigate towards the transmitted GPS waypoint. Once near the waypoint, the USV navigates to the target using its on-board camera and a computer vision model trained on the target, an orange life ring. All aspects of the autonomous SAR system have been tested, evaluated, and verified in an outdoor environment and this experimental data is presented and analyzed.

Keywords: Air Vehicle Operations, Integration, Path Planning
Abstract: Creating and executing missions for autonomous vehicles during search-and-rescue (SAR) operations is challenging for inexperienced operators and can become impossible under degraded communication conditions. Traditional ground control stations (GCS) require extensive workflows, mission changes can be slow to plan, and reliable internet connectivity cannot be assumed in certain field environments. The Lightweight Intent-Focused Engine for Guidance in Unmanned Autonomous Rescue Deployments (LIFEGUARD) is a self-contained, offline-capable GCS that converts spoken natural language intent into executable MAVLink missions for one or more autonomous agents. The system couples an offline speech-to-text (STT) engine (Vosk), a compact spaCy-based natural language understanding (NLU) stack with number normalization and GPS coordinate parsing, and a robust mission execution module built on PyMAVLink. Operators can issue voice commands, and LIFEGUARD extracts entities, generates and executes flight paths, and coordinates agents, all while including confirmation dialogs. We demonstrate software-in-the-loop (SITL) validation, showing multi-agent coordination in a multi-vehicle experiment executed without internet connectivity. The system achieves an end-to-end latency of <2 seconds from the push-to-talk (PTT) release to the initiation of mission upload, enabling responsive voice control of unmanned systems in time-critical SAR scenarios.

Keywords: Networked Swarms, Control Architectures
Abstract: In this paper, we address the distributed leader–follower formation control problem for multiple quadrotor aerial vehicles subject to hard amplitude constraints on the commanded translational velocities and yaw rate. Using a consensus-based formulation on the augmented pose, we design a low-complexity kinematic-level controller that enforces prescribed transient and steady-state performance for all local formation errors using only local neighbor information. To explicitly account for saturation in the embedded low-level flight controller, we introduce an adaptive prescribed performance mechanism that relaxes the performance bounds during saturation and recovers the nominal performance once feasibility is restored, thereby avoiding infeasible control demands and ensuring bounded closed-loop behavior. All closed-loop signals remain bounded and prescribed performance for the formation errors is guaranteed to the extent permitted by the imposed input constraints. Simulation results validate the effectiveness of the proposed approach.

Keywords: UAS Communications, Simulation, Reliability of UAS
Abstract: This paper presents a scalable airborne cellular relay system for emergency communication coverage, focusing on simulation-based coverage characterization and control-oriented system design. The proposed architecture adopts a control-oriented framework in which RF-IQ–based uplink sensing and multi-aircraft redundancy are defined as future extensions for adaptive and resilient operation in disaster scenarios. The relay payload is integrated into a certified Super SPIDI pod and is designed for integration with long-endurance aerial platforms. A dual-link architecture decouples the feeder link and service link, enabling frequency reuse and flexible backhaul configuration. Multi-gateway interference suppression using zero-forcing (ZF) was experimentally validated in an SDR environment, demonstrating significant error vector magnitude (EVM) improvement under cochannel interference conditions. Simulation-based evaluations were conducted to characterize reference signal received power (RSRP) distributions under helicopter-assisted deployment geometries and to identify practical deployment constraints, including antenna actuation limits. The results establish quantitative coverage trends with respect to altitude and horizontal distance and provide a structured validation framework toward future field experiments. The proposed system emphasizes practical airborne integration and provides a foundation for the deployment of resilient airborne communication systems in large-scale disaster response scenarios.

Keywords: UAS Communications, Perception and Cognition, Smart Sensors
Abstract: Unmanned Aerial Vehicles (UAVs) are popular robotic platforms nowadays and they are gradually entering our everyday lives. For normal operation, a two-way communication link between a UAV and a ground station, has to be established. While radio-frequency communication is a simple and common option, it suffers from major drawbacks, such as interference and multi-path errors. Moreover, it is prone to spoofing/jamming attacks which might have catastrophic consequences in densely-populated or sensitive areas. To overcome these limitations, in this paper, we propose a compact LED-event camera system for fast optical communication via a new optimized dynamic N-pulse protocol. A learning-based method guarantees the robust detection and tracking of the active markers and it makes the application of standard 3D pose estimation algorithms possible. The effectiveness of the proposed system is demonstrated via hardware experiments with a DJI Matrice 600 Pro hexarotor and a Prophesee Gen. 4.1 HD camera.

Keywords: UAS Communications, UAS Applications, Simulation
Abstract: Unmanned Aerial Vehicles (UAVs) are increasingly employed in long-distance urban monitoring and inspection missions that require reliable high-throughput wireless connectivity for real-time video transmission. Evaluating such communication strategies in large-scale urban environments is costly and difficult to reproduce experimentally, making simulation a fundamental tool. However, many existing platforms treat UAV mobility and wireless communication independently, neglecting their strong coupling under frequent handovers and data-intensive traffic. This work presents the modeling and evaluation of an SNR- and bandwidth-aware handover strategy for UAV-based urban monitoring. The proposed approach prioritizes throughput while preserving link robustness by triggering handovers only when the received SNR falls below a predefined threshold. Simulation results in representative urban scenarios demonstrate improved link stability, controlled packet loss, and sustained data rates during mission execution.

Keywords: Swarms, Path Planning, Manned/Unmanned Aviation
Abstract: Rapid acquisition of situation awareness and the execution of complex, multi domain missions are critical in disaster response operations. The Swarming project, a research effort at the German Aerospace Center (DLR), addresses this challenge, where a fleet of heterogeneous unmanned aircraft and ground systems operated in swarms provide critical information from a disaster site in real-time. The swarms are coordinated in a task-based manner from one central ground control station. Furthermore, a manned helicopter serves as a mothership for unmanned aircraft that can be dropped in flight and autonomously integrated into a swarm. This paper describes the components that were developed, integrated into a system-of-systems and demonstrated within the scope of a simulation campaign using a multi-agent simulation system. An outlook on the next steps for further exploitation of the Swarming platform is given.

Keywords: Networked Swarms, Swarms, Autonomy
Abstract: This work presents a human-in-the-loop framework for coordinating multi-UAV teams in disaster environments, where a large language model serves as a natural-language interface between a human operator and a swarm of UAVs operating under short-range communication constraints. The system is demonstrated in a mission scenario in which a ground robot explores an environment while UAVs act as communication relays to maintain connectivity between the robot and a base station. The operator provides high-level guidance that complements an autonomous connectivity-and-perching controller. The framework is evaluated in multiple simulated environments using connectivity and mission-duration metrics, together with interaction measures such as command-to-execution delay. Experiments show that trained user interaction can outperform the fully autonomous baseline in multiple environments.

Keywords: Risk Analysis, Path Planning, Regulations
Abstract: This paper proposes a path planning algorithm that explicitly accounts for ground risk constraints as directly expressed by the SORA risk assessment method, considered in UE regulations on UAVs. Two criteria are handled: one related to the Ground Risk Class of the operation, and another on containment requirements related to the risk of fly-away and crash over the Adjacent Area of the operation. A filtering process of the population density map is proposed along with a path planning algorithm that automatically and iteratively adjusts the trade-off between the length of the path and the ground risk criterion. The proposed algorithm is illustrated on an UAV operation close to areas with high population densities.

Keywords: Risk Analysis, Manned/Unmanned Aviation, Levels of Safety
Abstract: The integration of Uncrewed Aerial Vehicles (UAVs) into non-segregated airspace presents a critical safety challenge, requiring approval mechanisms that are both rigorous and operationally flexible. This paper proposes a hybrid risk assessment framework for operations near airways that synthesizes deterministic geometric constraints with probabilistic collision risk modeling. The assessment logic operates on a two-tier hierarchy: first, a strict geometric veto prohibits operations within the 3 sigma deviation safety bounds of crewed airways; second, it evaluates the quantitative collision (NMAC) risk against a Target Level of Safety (TLS) threshold. We assess this framework using a cluster-based air traffic model derived from real-world traffic data in a terminal area. The results validate that while geometric conflicts require spatial replanning, risk-based conflicts can often be resolved through temporal scheduling, maximizing airspace access without compromising safety.

Keywords: Fail-Safe Systems, Micro- and Mini- UAS, Risk Analysis
Abstract: Indoor micro-aerial vehicles (MAVs) are increasingly considered for tasks that bring them in close proximity to people, yet practitioners lack a practical way to tune motion limits to measured impact risk. We present an end-to-end, open toolchain that turns benchtop impact tests into deployable safety governors for drones. First, we detail a compact, replicable impact rig and protocol that captures force–time profiles across drone classes and contact surfaces. Second, we provide data-driven fits that map pre-impact speed to impulse and contact duration, enabling direct computation of speed bounds for a target force limit. Third, we release scripts and a ROS2 node that enforce these bounds online and log compliance, with hooks for facility-specific policies. We target indoor MAV platforms with protective cages, focusing on bounding blunt contact forces of the airframe rather than propeller-strike hazards. We validate runtime enforcement in a closed-loop PX4-SITL/Gazebo setup (rotor-on dynamics), while physical rotor-on impact testing is left for future work. The contribution is a practical bridge: from measured impacts to runtime limits, with sharable datasets, code, and a repeatable process that teams can adopt to certify indoor MAV operations near humans.

Keywords: Risk Analysis, Manned/Unmanned Aviation, UAS Applications
Abstract: UAV intrusion events pose significant safety risks to airport operation due to their high uncertainty, limited observability and potential operational disruptions. This paper proposes an integrated trajectory uncertainty prediction and collision risk assessment framework for UAV intrusion scenarios in airport terminal airspace. A probabilistic trajectory prediction model is developed to estimate both UAVs and manned aircraft future positions and their associated uncertainties. Based on these heterogeneous trajectory predictions, a collision risk assessment method is established to derive the lower bound of time-to-collision and upper bound of collision probability in both nominal and worst-case intrusion scenarios. Quantitative risk indicators are also provided to support maneuvering and countermeasure decisions during critical flight phases. Results demonstrate that the proposed framework can provide accurate collision risk estimates, supporting timely decision-making under representative UAV intrusion events.

Keywords: Manned/Unmanned Aviation
Abstract: With the widespread application of unmanned aircraft systems (UAS) across various fields, non-cooperative UAS, characterized by frequent maneuvers and strong motion uncertainty, pose potential risks to airspace safety, making their trajectory prediction particularly challenging. To address these challenges, this paper proposes a hybrid trajectory prediction framework for non-cooperative UAS, which integrates LSTM-Gaussian with an interacting multiple model (IMM). By introducing a Gaussian loss function, the LSTM-Gaussian model is able to characterize prediction uncertainty while outputting trajectory mean estimates, thereby enabling confidence interval estimation for future states. To effectively handle diverse flight motion patterns, the IMM structure models three typical motion modes, including constant velocity, constant acceleration, and coordinated turn. An adaptive fusion strategy is designed to organically combine the nonlinear modeling capability of LSTM with the numerical stability and physical interpretability of IMM. Experimental results demonstrate that the proposed method achieves higher prediction accuracy and robustness across different motion scenarios, with particular advantages in short-term prediction, providing a reliable and effective solution for trajectory prediction of non-cooperative UAS under complex maneuvering conditions.

Keywords: Manned/Unmanned Aviation, See-and-avoid Systems, Simulation
Abstract: Various probabilistic and deterministic methods have been developed and extensively validated in simulation environments to enhance safety in hybrid manned-unmanned airspace. However, reliance on simulation alone limits real-world assessment of algorithm reliability. This work closes a critical simulation-to-flight validation gap for right-of-way (RoW) logic by experimentally evaluating a morphing potential field algorithm in representative encounters and quantifying the minimum detection distance required for a fixed-wing UAS to maintain well-clear from a crewed aircraft. Flight tests were conducted using a general aviation Cessna 172 Skyhawk and a fixed-wing UAS, both instrumented with ADS-B to enable cooperative DAA functionality. Seventeen flight test scenarios were conducted, systematically varying the relative heading and speed to generate a diverse set of encounters and to map well-clear outcomes as a function of detection range. Results indicate that simulations accurately predicted the UAS's ability to maintain the required 2000-ft right-of-way separation in most cases, with 88% of flight test outcomes aligning with simulation-based well-clear classifications. Taken together, the experimentally observed detection range thresholds and the measured simulation-to-flight mismatch demonstrate that high-fidelity nonlinear simulation provides strong predictive capability for DAA performance and can effectively guide design and pre-flight screening. However, the observed sensitivity to nonlinear right-of-way logic, encounter phasing, and unmitigated collision distance highlights the necessity of at least a detection range of 3,962-m or greater in the tested ranges to meet well-clear requirements when utilizing the morphing potential field algorithm.

Keywords: Reliability of UAS, Risk Analysis, Fail-Safe Systems
Abstract: Onboard UAV trajectory prediction for safety- critical missions requires not only accurate forecasts but also de- ployment-oriented epistemic uncertainty under strict millisec- ond-level inference budgets. Existing evaluations are often budget-misaligned, comparing single-forward predictors with multi-forward baselines without normalizing inference cost. In this paper, a budget-aligned evaluation protocol is established based on the forward-pass budget, floating-point operations (FLOPs), and CPU-proxy latency. Outage-segment trajectory- level survival is introduced to assess system-level safety. Under the strict single-forward constraint, deep deterministic uncer- tainty (DDU) is adapted to time-series regression by stabilizing the feature space with spectral normalization, modeling feature density via a Gaussian mixture model (GMM), and mapping density scores to continuous epistemic variance through isotonic calibration. Stress tests on a physics-intensity grid demonstrate that the proposed approach retains safety under severe runtime drift. In the HighDyn_LongOut scenario, a trajectory-level sur- vival probability of 0.92 is achieved, while a five-member deep ensemble baseline yields 0.45. The proposed post-processing in- troduces only about 8.5% overhead and provides approximately 4.6 times speedup relative to the ensemble baseline under CPU- proxy evaluation, supporting onboard-feasible epistemic uncer- tainty estimation within a millisecond-level budget under the evaluated simulation setting.

Keywords: Reliability of UAS, Perception and Cognition, Autonomy
Abstract: Autonomous drones are used in a wide range of applications such as package delivery, search and rescue, surveillance, and other special operations. Most of autonomous drones are equipped with object detection models to identify generic objects or models specially trained for objects such as humans, vehicles, or other drones in a swarm. One of the main challenges of detecting objects during adverse weather conditions is when the drone is not capable of acquiring undistorted images under the glare haze produced. As object detection is paramount in drone vision, the ability to use pretrained detection models with its full potential even during such conditions is indispensable for full autonomy. To improve the visibility in bad weather, there are various kinds of approaches have been taken, including physically grounded statistical methods and deep learning based techniques. However, adding heavy processing to the detection pipeline is undesirable, as it adds substantial latency to every frame. To address this, we developed method to gauge the extent of the haze real-time and adjusting the de-hazing magnitude without distorting the image, or bypass de-hazing altogether when unnecessary. Our method consist of analyzing the light and dark channels immediately after the image acquisition or right after the object detection to produce a reliable estimate of the haze level providing a critical foundation for controlled de-hazing, adaptive decision making, and safe autonomous drone operation across varying haze conditions.

Keywords: Reliability of UAS, UAS Applications, Fail-Safe Systems
Abstract: This paper presents a transfer learning framework for cross-platform propeller damage regression in multirotor UAVs. A neural network is trained in a source domain to estimate asymmetric propeller tip damage from spectral features extracted from inertial and control signals. The model is then deployed on a target vehicle with different size and propulsion characteristics, introducing domain shift. We evaluate three adaptation strategies under realistic deployment constraints: physics-based spectral scaling (zero-shot), affine output calibration, and few-shot fine-tuning. Experiments on two quadrotor platforms show that spectral scaling provides modest gains in strictly zero-shot settings, while few-shot fine-tuning substantially reduces estimation error. Interpolation and extrapolation analyses indicate that fine-tuning improves accuracy near supervised damage levels, whereas spectral scaling produces more uniform behavior across unseen magnitudes. These results characterize practical trade-offs when deploying regression-based fault estimation models across heterogeneous UAV platforms under limited supervision.

Keywords: Sensor Fusion, Fail-Safe Systems, UAS Applications
Abstract: This work introduces a real-time, lightweight system deployed onboard unmanned aerial vehicles (UAVs) that achieves the two-fold objective of state classification and trajectory prediction by integrating a Transformer-based physics-informed learning model with an extended Kalman filter (EKF). The proposed framework employs a two-stage sensor fusion mechanism that combines a physics-informed Transformer with an EKF-based measurement augmentation. In the first stage, an EKF fuses onboard sensor readings together with vision-derived measurements (optical flow measurements) and external sensor information (such as weather station data of wind speed and direction) to construct an augmented and denoised input measurement vector. In the second stage, the resulting filtered state estimates are fed into a physics-informed Transformer network that exploits shared feature learning to jointly enhance robustness and improve state classification and trajectory forecasting performance in Global Navigation Satellite System (GNSS)-denied environments. The proposed EKF-assisted AI framework is trained and validated on a custom dataset of UAV trajectories collected across various outdoor environments, and its prototype implementation is thoroughly assessed and validated in real-world outdoor experiments.

Keywords: Fail-Safe Systems, Multirotor Design and Control, Reliability of UAS
Abstract: The operation of heavy hybrid Unmanned Aerial Vehicles (UAVs) in populated areas demands robust Fault Tol- erant Control (FTC) strategies to ensure safe landings following actuator failures. This paper presents the first successful real-life demonstration of stable relaxed hover on a large hybrid drone, the Variable Skew Quad Plane (VSQP), subject to the total loss of control authority in two opposing multirotor motors while in forward-flight mode. A modified Incremental Nonlinear Dynamic Inversion (INDI) controller is derived to regulate the vehicle’s principal axis of rotation. The framework is extended to actively incorporate aerodynamic control surfaces (flaperons), which are shown to significantly improve the rejection of unmodelled aerodynamic and reaction moments. Furthermore, alongside a standard position controller, a novel pusher-motor guidance strategy is proposed and flight-tested. By exploiting the pusher motor the control-authority requirements on the primary stabilization actuators are reduced. The controller achieves fully autonomous tracking of a square trajectory. These results confirm the efficacy of INDI for executing re- laxed hover on large quad-planes equipped with aerodynamic surfaces, and demonstrate the benefits of leveraging the full suite of hybrid UAV actuators.

Keywords: Airworthiness, Reliability of UAS, Levels of Safety
Abstract: Propeller damages can have a substantial impact on the performance of the powertrains of unmanned aerial vehicles (UAVs). In the case of significant damage, the consequences may be critical with regard to flight safety. To prevent taking off with damaged propellers, the integrity of the installed propellers is assessed during pre-flight checks. This paper explores the potential for automating this inspection by using computer vision models trained on UAV image data. Following an initial data collection phase, computer vision models are implemented and their performance is assessed. The approach demonstrates that the implemented models successfully detect damaged propellers with high accuracy, especially in the case of severely damaged propellers, which are most safety-critical.

Keywords: UAS Applications, Perception and Cognition, UAS Testbeds
Abstract: With the rapid proliferation of unmanned aerial systems (UAS) and growing interest in onboard computation, 3D target localization has emerged as a fundamental challenge across many applications. This work demonstrates the feasibility of a lightweight method for estimating the 3D locations of stationary targets using ray-ray intersections combined with efficient filtering techniques. The proposed approach enables localization of multiple targets from asynchronous and noisy observations, even in the presence of false positives, using data from a single UAS. The method relies only on sensors commonly available on autonomous UAS, namely a monocular camera and a position estimate, making it well-suited for smaller, low-cost platforms. Experimental results showcase localization accuracy comparable to, and in some cases exceeding, existing approaches, while maintaining low computational complexity.

Keywords: Perception and Cognition, Autonomy, UAS Applications
Abstract: Autonomous recovery of Unmanned Aerial Vehicles (UAVs) in maritime environments is challenged by dynamic ship motion, texture-poor surroundings, and vulnerabilities inherent to Global Navigation Satellite Systems (GNSS). To enable robust UAV state estimation relative to a naval vessel in GNSS-denied environments, this paper presents a novel vision-based pose estimation framework leveraging Feature 3D Gaussian Splatting (3DGS). Adopting STDLoc's sparse pose estimation framework, our approach bypasses computationally expensive dense rendering in favor of direct sparse 2D-3D matching. We extract 2D features from live monocular imagery using SuperPoint and match them against a compact set of view-consistent 3D Gaussian landmarks distilled via a Matching-Oriented Sampling strategy. The 6-DoF camera pose is then robustly recovered using a Perspective-n-Point (PnP) solver augmented with Locally Optimized RANSAC (LO-RANSAC). Experimental validation on a 1/24-scale naval Yard Patrol (YP) craft demonstrates centimeter-level translational accuracy (1.66 cm lateral, 2.28 cm vertical mean absolute errors) and near-degree rotational accuracy (0.90° roll, 1.17° yaw). This method eliminates the need for physical fiducial markers or precise CAD models, offering a scalable and highly accurate relative localization solution for autonomous maritime operations.

Keywords: Sensor Fusion, Perception and Cognition, Navigation
Abstract: Autonomous UAV operations on ships require reliable vision-based relative pose estimation, yet at-sea validation is costly, weather-dependent, and risky. This paper presents a hardware-validated vision-in-the-loop framework that enables fully autonomous indoor flight while emulating photorealistic maritime environments. Rendered maritime views are processed onboard by a deep transformer-based monocular pose estimator. Delayed vision measurements are fused with high-rate IMU data using a delayed Kalman filter to provide consistent state estimates for geometric control. The system captures critical embedded effects, including perception latency, asynchronous updates, and computational constraints, that are absent in pure simulation. Autonomous takeoff, trajectory tracking, and landing experiments demonstrate stable closed-loop flight. The results establish a safe and hardware-realistic intermediate stage for developing maritime UAV autonomy prior to shipboard deployment.

Keywords: Risk Analysis, Autonomy, Airworthiness
Abstract: As Mission-Level Autonomous Systems (MAS) for aviation systems become ready for deployment, safety is of paramount importance, especially in conditions where these systems interact with humans. Compounding this general challenge is the need for rigorous certification and safety guarantees that are hallmarks of the aviation community; namely, the exploration of an intractably large sample space is an important consideration when characterizing likelihood of hazards. This brings up the question of how can aviation safety practitioners capture and characterize risk when the computational budget for computer simulation is limited, and the real-world data is sparse while still ensuring mission success? One approach in the computer simulation literature is the development of surrogate models trained on limited datasets to perform large state exploration. This work expands on that approach by using Gaussian Process Regression, a supervised learning approach, to build a surrogate for further safety analysis and Operational Design Domains to precisely define the sampling space. This paper presents a methodology to build a model specifically to estimate risk for MAS by using a large computer simulation in support of safety certification.

Keywords: Autonomy, Sensor Fusion, Micro- and Mini- UAS
Abstract: This paper tests and evaluates the performance of visual-inertial odometry (VIO) under adverse visual factors within the context of autonomous proximity flight. VIO is widely used in UAV autonomy for micro air vehicles (MAVs) due to its cost-effectiveness and the small size of required sensor. Since VIO relies on a camera, visual factors can significantly impact its performance; however, there is a general lack of public VIO flight datasets that contain adverse visual factors. Here we describe the use of a MAV platform to collect real-world visual-inertial datasets that contain adverse visual factors– for example, moving objects, reflections, low light, featureless environments, and lens contamination, along with the ground truth pose of the MAV. However, some visual factors, such as f ire, dust, and smoke, are challenging or hazardous to reproduce in a controlled environment. To produce datasets containing these particular visual factors, we combine and augment the real-world images with a computational simulation. The real and augmented datasets along with ground-truth data are used to evaluate VIO robustness through a combination of practical performance metrics.

Keywords: UAS Testbeds, Manned/Unmanned Aviation, Energy Efficient UAS
Abstract: This project investigates the aerodynamic optimization of a flying-wing design using the Cross Entropy Method (CEM), a global stochastic optimization technique. The objective was to identify an optimal planform that maximizes aerodynamic performance, defined as a weighted combination of lift and lift-to-drag ratio, while satisfying stability constraints for multi-role intelligence, surveillance and reconnaissance (ISR) and strike applications. A vortex lattice method (VLM) was used to compute aerodynamic forces in inviscid flow, and Trefftz-plane analysis was employed to calculate induced drag. Design variables included local sweep, twist, taper ratio, and dihedral angles of both the main wing and tip segment. Constraints were enforced via a quadratic penalty formulation to maintain a minimum static margin, ensuring longitudinal stability. The CEM iteratively refined a probability distribution over the design space, converging to a planform that demonstrated theoretical improved aerodynamic performance over baseline and compliance with geometric and stability constraints. The final design indicated reduced induced drag and enhanced mission-specific aerodynamic efficiency. Flight test validation attempts resulted in rapid-unscheduled disasssembly. Further flight test validation is required.