Keywords: Control Architectures, Reliability of UAS
Abstract: Many unmanned aerial vehicles (UAVs) can remain aerodynamically flyable after sustaining structural or control surface damage, yet insufficient robustness in conventional autopilots often leads to mission failure. This paper proposes a robust adaptive sliding mode controller (RASMC) for fixed-wing UAVs subject to aerodynamic coefficient perturbations and partial loss of control surface effectiveness. A damage-aware flight dynamics model is developed to systematically analyze the impact of such impairments on the closed-loop behavior. The RASMC is designed to ensure reliable tracking and stabilization, while a gain adaptation law maintains low control effort under nominal conditions and increases the gains as needed in the presence of aerodynamic damage. Lyapunovbased stability guarantees are derived, and assumptions on admissible uncertainty bounds are formulated to characterize the limits within which closed-loop stability and performance can be ensured. The proposed controller is implemented within an existing UAV autopilot framework, where outer-loop guidance and speed control modules provide reference commands to the RASMC for attitude stabilization. Simulations demonstrate that, despite significant damage, all closed-loop states remain stable with bounded tracking errors.

Keywords: Simulation, Reliability of UAS, Control Architectures
Abstract: Co-design optimisation of autonomous systems has emerged as a powerful alternative to sequential approaches by jointly optimising physical design and control strategies. However, existing frameworks often neglect the robustness required for autonomous systems navigating unstructured, real-world environments. For agile Unmanned Aerial Vehicles (UAVs) operating at the edge of the flight envelope, this lack of robustness yields designs that are sensitive to perturbations and model mismatch. To address this, we propose a robust co-design framework for agile fixed-wing UAVs that integrates parametric uncertainty and wind disturbances directly into the concurrent optimisation process. Our bi-level approach optimises physical design in a high-level loop while discovering nominal solutions via a constrained trajectory planner and evaluating performance across a stochastic Monte Carlo ensemble using feedback LQR control. Validated across three agile flight missions, our strategy consistently outperforms deterministic baselines. The results demonstrate that our robust co-design strategy inherently tailors aerodynamic features, such as wing placement and aspect ratio, to achieve an optimal trade-off between mission performance and disturbance rejection.

Keywords: Control Architectures, Aerial Robotic Manipulation, UAS Applications
Abstract: This work presents a simulation-based comparative robustness analysis of Incremental Nonlinear Dynamic Inversion (INDI) and Nonlinear Dynamic Inversion augmented with a nonlinear disturbance observer (NDI+NDO) for fully actuated aerial robots. A systematic simulation campaign across representative operating scenarios is conducted, where we compare tracking performance, robustness, control effort, under parametric variations, external disturbances, and measurement noise. Results show that INDI demonstrates stronger robustness in several model-mismatch and combined-stress cases, while NDI+NDO primarily matches nominal performance but exhibits greater sensitivity under several non-ideal conditions. These findings provide practical guidance on the relative strengths and limitations of incremental and observer-based inversion strategies for aerial robotic applications.

Keywords: Control Architectures, Micro- and Mini- UAS, UAS Applications
Abstract: Unmanned Aerial Vehicles (UAVs) are susceptible to losing stability and crashing while operating in a turbulent environment, therefore the safety and reliability of UAV control in such environments is a matter of great concern. Devising a robust control algorithm to reject disturbances is challenging due to the highly nonlinear nature of wind dynamics, and modeling the dynamics using analytical techniques is not straightforward. While traditional techniques using disturbance observers and classical adaptive control have shown some progress, they are mostly limited to relatively non-complex environments. On the other hand, learning-based approaches are increasingly being used for modeling of residual forces and disturbance rejection; however, their generalization and interpretability is a factor of concern. To this end, we propose a novel integration of data-driven system identification using Sparse Identification of Nonlinear Dynamics (SINDy) with a Recursive Least Square (RLS) adaptive control to adapt and reject wind disturbances in a turbulent environment. We tested and validated our approach on Gazebo harmonic environment and on real flights with wind speeds of up to 2 m/s from four directions, creating a highly dynamic and turbulent environment. The drone tracked complex trajectories like circular, lemniscate, and spiral with a speed of up to 0.89 m/s. Adaptive SINDy outperformed the baseline PID and INDI controllers on several trajectory tracking error metrics without crashing. A root mean square error (RMSE) as low as 17.6 cm and 12.2 cm, and a mean absolute error (MAE) of 13.7 cm and 10.5 cm were achieved on circular and lemniscate trajectories, respectively. The validation was performed on a very lightweight Crazyflie drone under a highly dynamic environment.

Keywords: Autonomy, Control Architectures, Navigation
Abstract: Wind disturbances remain a key barrier to reliable autonomous navigation for lightweight quadrotors, where rapidly varying airflow can destabilize both planning and tracking. This paper introduces GustPilot, a hierarchical wind resilient navigation stack in which a deep reinforcement learning (DRL) policy generates inertial-frame velocity references for gate traversal, while a geometric Incremental Nonlinear Dynamic Inversion (INDI) controller provides low-level tracking with fast residual disturbance rejection. The INDI layer uses incremental feedback on both specific linear acceleration and angular acceleration/rate, relying on onboard sensor measurements to reject wind disturbances during execution. Robustness is achieved through a two-level strategy: wind-aware planning learned via fan-jet domain randomization during training and rapid execution-time disturbance rejection by the INDI tracking controller. We evaluate GustPilot in real flights on a 50 g quadrotor platform against a DRL–PID baseline across four scenarios ranging from no-wind to fully dynamic conditions with a moving gate and a moving disturbance source. Despite being trained only in a minimal single-gate/single-fan setup, the policy generalizes to more complex environments with up to six gates and four fans without retraining. Across 80 experiments, DRL–INDI achieves an average Overall Success Rate (OSR) of 94.6%, compared with 36.0% for DRL–PID, reduces tracking Root Mean Square Error (RMSE) by up to 50%, and sustains speeds up to 1.34 m/s under wind disturbances up to 3.5 m/s. These results demonstrate that combining DRL-based velocity planning with structured INDI disturbance rejection provides a practical approach to wind-resilient autonomous flight.

Keywords: Payloads, Control Architectures, UAS Applications
Abstract: Some aerial applications require controllers that are robust to system uncertainties and external disturbances. When an aircraft carries a mass that can vary over time, this can introduce or increase uncertainties in the system and affect its performance. Similarly, if this aircraft carrying a load is exposed to gusts of wind, this will also degrade its performance. In literature, robust controllers are proposed to mitigate these effects, nevertheless, their application in real-time experiments is sometimes hard and requires high-gain values to ensure the expected robustness. In this work, a hybrid robust control architecture based on a nominal controller with two compensation mechanisms is proposed. The compensation mechanism is composed by (i) an adaptive law proposal to compensate uncertainties into the system as mass variations and (ii) a Backpropagation Neural Network (BNN) to estimate the external disturbances during flight. The Lyapunov analysis is used to demonstrate that the errors of the system in closed-loop are Uniformly Ultimately Bounded (UUB). To corroborate the performance of the proposed strategy, several real-time flight tests are carried out, comparing the nominal controller with the adaptive baseline against its BNN-augmented version.

Keywords: Control Architectures, Simulation
Abstract: This paper presents a residual-enhanced Koopman-based Model Predictive Control (RKMPC) framework for quadrotor trajectory tracking under parametric and unmodeled uncertainties. Building upon a quasi-linear lifted representation obtained through Koopman operator theory, a low-dimensional linear MPC formulation is preserved while augmenting the prediction model with a learned residual correction. The residual term is trained offline using simulation data and evaluated outside the quadratic program at each sampling instant, where it is injected as a measured disturbance. This strategy maintains the convex QP structure and computational efficiency of the baseline Koopman MPC while compensating prediction bias induced by modeling errors. Closed-loop simulations are conducted under both nominal conditions and uncertain plant parameters. The proposed RKMPC significantly improves tracking performance and robustness against parameter mismatches between the true and simulated quadrotor plant. The results highlight the robustness–optimality trade-off between nominal model-based control and residual-enhanced predictive control, demonstrating that learned residual compensation enhances robustness without increasing online computational complexity.

Keywords: Multirotor Design and Control, Control Architectures, Aerial Robotic Manipulation
Abstract: In this work, we present a novel optimization-based input allocation strategy for fully-actuated multirotor aerial vehicles with N propellers that explicitly respects box constraints on the control inputs in both positive and negative actuation values while accounting for a deadband around zero. The resulting optimization problem belongs to the fundamental class of NP-hard global optimization problems, which can be recast as a Mixed-Integer Linear Program (MILP). We show that feasible solutions to this formulated MILP can efficiently be computed using a standard branch-and-bound algorithm. Building on that algorithm, the proposed method also robustly handles cases in which the desired allocation is infeasible by introducing several fallback instances. The method is experimentally validated on an octo-rotor platform, demonstrating the effectiveness of the proposed approach and its superior performance compared to the conventional QP-based allocation method.

Keywords: Multirotor Design and Control, Path Planning, Payloads
Abstract: This paper develops a learning-based tracking controller for cooperative transport of a rigid payload by multiple aerial manipulators under rigid grasp constraints. A unified geometric model yields a coupled agent–payload differential–algebraic system capturing contact wrenches and internal-force redundancy. A leader generates a desired payload wrench from geometric tracking errors, and follower agents realize it via constraint-consistent wrench allocation.

Model uncertainties and disturbances are compensated using Gaussian Process (GP) regression. High-probability GP error bounds are embedded in the controller through a GP feedforward term and geometric feedback. A Lyapunov analysis guarantees uniform ultimate boundedness of the payload tracking errors with high probability, with a bound that scales with the GP predictive uncertainty.

Keywords: Control Architectures, Multirotor Design and Control, Aerial Robotic Manipulation
Abstract: This paper presents a sensitivity-based tube Nonlinear Model Predictive Control (NMPC) framework for cooperative aerial chains under bounded parametric uncertainty. We consider a planar two-vehicle chain connected by rigid links, modeled with input-rate actuation to enforce slew-rate and magnitude limits on thrust and torque. Robustness to uncertainty in link mass, length, and inertia is achieved by propagating first-order parametric state sensitivities along the horizon and using them to compute online constraint-tightening margins. We robustify an inter-link separation constraint, implemented via a smooth cosine embedding, and thrust-magnitude bounds. The method is implemented in MATLAB and evaluated with boundary-hugging maneuvers and Monte-Carlo uncertainty sampling. Results show improved constraint margins under uncertainty with tracking performance comparable to nominal NMPC.

Keywords: Multirotor Design and Control, Autonomy, Simulation
Abstract: Fully actuated omnidirectional UAVs enable independent control of forces and torques along all six degrees of freedom, broadening the operational envelope for agile flight and aerial interaction tasks. However, conventional control allocation methods neglect the asymmetric dynamics of the onboard actuators, which can induce oscillatory motor commands and degrade trajectory tracking during dynamic maneuvers. This work proposes a receding-horizon, actuation-aware allocation strategy that explicitly incorporates asymmetric motor dynamics and exploits the redundancy of over-actuated platforms through nullspace optimization. By forward-simulating the closed-loop system over a prediction horizon, the method anticipates actuator-induced oscillations and suppresses them through smooth redistribution of motor commands, while preserving the desired body wrench exactly. The approach is formulated as a constrained optimal control problem solved online via Constrained iterative LQR. Simulation results on the OmniOcta platform demonstrate that the proposed method significantly reduces motor command oscillations compared to a conventional single-step quadratic programming allocator, yielding improved trajectory tracking in both position and orientation.

Keywords: Control Architectures, Multirotor Design and Control
Abstract: This paper presents a geometric adaptive control framework to control fully actuated aerial vehicles that are subjected to variations in mass, CoG and moment of inertia due to unknown payload events. The controller is formulated on SE(3) and uses Lie-algebra errors in se(3) to achieve global, coordinate-free pose tracking. An adaptation law is derived for online estimation of the generalized inertia parameters with Lyapunov stability guarantees. The approach is validated in MATLAB simulations over multiple trajectories with an unknown initial payload and an abrupt payload drop. Results demonstrate consistently high SE(3) tracking performance, rapid recovery after parameter changes, and substantial improvement over a non-adaptive baseline. Mass estimation is reliable, CoG estimation achieves partial convergence and improves with higher CoG adaptation gains, while inertia parameters show poor convergence due to insufficient regressor excitation, although this does not cause any system instability.

Keywords: Multirotor Design and Control, Control Architectures
Abstract: Fully actuated multirotor platforms decouple translational force generation from vehicle attitude, enabling independent control of position and orientation and shifting performance limitations from attitude authority to actuator dynamics and control effectiveness. This paper compares a model-based nonlinear dynamic inversion controller (geometric NDI) with a sensor-based incremental dynamic inversion controller (INDI) on a fixed-tilt fully actuated hexarotor. Both controllers share an identical outer-loop structure and are both executed at 500~Hz; therefore, performance differences can be attributed primarily to the inversion strategy. Controller performance is evaluated in five experiments covering attitude step tracking under nominal conditions and under a 50% mismatch in the rotor force coefficient, hover disturbance rejection under an external lateral load, waypoint tracking in the presence of wind gust disturbances, reduced control frequency, and injected sensor degradation. The results show that INDI offers clear advantages under parameter mismatch, gust disturbances, and sensor degradation, and maintains lower position errors across the controller-frequency sweep. However, its advantages are not universal: geometric NDI yields better attitude tracking at reduced control frequencies. To the authors’ best knowledge, this work presents the first experimental validation of a full pose tracking INDI controller with decoupled translational and rotational dynamics. These findings highlight the trade-off between measurement-based and model-based inversion for robust control and rapid deployment of fully actuated UAVs.

Keywords: Multirotor Design and Control, Control Architectures
Abstract: In this paper, we propose a geometric control scheme for multirotors with tiltable rotors. The control scheme is based on the well-known inner-outer loop structure, allowing the system to track both position and orientation references while avoiding local coordinate parameterizations, thereby preventing singularity issues. Moreover, it leverages thrust vectoring to enable level flight during trajectory tracking without requiring the entire vehicle to tilt to generate lateral forces, providing advantages for passenger transportation, infrastructure inspection, and, more generally, tasks involving interaction with the environment. Simulations conducted on a hexarotor show that the proposed control scheme enables level flight, is robust with respect to constant external disturbances such as steady wind, and can easily be tuned to accommodate rotor inefficiencies at allocation level.

Keywords: Path Planning, Navigation, Autonomy
Abstract: Omnidirectional drones offer a unique advantage over classical multirotor UAVs by enabling independent tracking of both three-dimensional position and orientation. This added flexibility makes trajectory planning significantly more challenging than planning for classical drones. In this paper, we present a trajectory planning framework tailored for omnidirectional UAVs operating in unknown, cluttered environments. The vehicle is equipped with a forward-facing RGB-D camera and an IMU, and builds an RTAB-Map with loop closures during flight. Our method incrementally identifies feasible subgoals from the explored free space and generates full-pose safe trajectories—covering both position and orientation—to reach them. This process is repeated until the global target becomes accessible. The proposed approach is validated in a high-fidelity physics simulator, using our omnidirectional drone, within an environment that requires precise navigation through tight passages and complex orientations. The planner’s effectiveness is demonstrated in scenarios that are otherwise infeasible for conventional UAVs.

Keywords: Path Planning, UAS Applications
Abstract: The multi-agent routing is the first stage of an ensemble mission planning and execution pipeline required to control a multi-agent system in an environment filled with obstacles while ensuring deconfliction between the agents. The concept of the presented multi-agent routing aims at allocating mission goals in an optimized order to the agents. Moreover, it plans obstacle-free corridors through the environment represented by an n-dimensional tree. The multi-agent routing obtains the corridors using A* search algorithm. We propose a new approach to calculating the edge weights for the A* search algorithm that significantly shortens the overall corridor length and improves flight safety. The advantage of the proposed method is studied in a realistic urban environment, and its performance is compared with that of the standard approach, which uses Euclidean distance between the node center points as the edge weight.

Keywords: Path Planning, Swarms, Control Architectures
Abstract: Path and trajectory planning for multi-agent systems is computationally heavy when vehicle dynamics are involved. If a centralized setup is used, the computational burden scales exponentially with the number of agents due to dependent decisions and interactions between them. To dampen this effect the classical idea is decentralization. However, optimality gets lost and in a model predictive control setup major replanning effects occur if no consensus is enforced. To overcome these issues, we propose another planning layer providing decisions and additional information to significantly simplify the dynamics based lower-level optimization.

Keywords: Path Planning, Multirotor Design and Control
Abstract: The agility of quadcopters enables diverse autonomous applications, but rapid trajectory planning in cluttered environments remains challenging due to strict payload constraints on sensing and computation. This paper presents a computationally efficient, memoryless local path planner for navigating quadcopters over obstacles using limited onboard resources. Operating at 20Hz in a receding horizon fashion, the planner relies solely on the current vehicle state and the latest depth measurements. We introduce a plane-based spatial partitioning method that accelerates trajectory collision checking and selects optimal motion primitives to maximize mission velocity. Our algorithm reduces collision-checking duration to approximately 25% of the prior pyramid-based method while maintaining comparable mission completion behavior. The system is validated through simulation and physical experiments, demonstrating safe and efficient navigation toward designated goals above obstacles.

Keywords: Path Planning, Autonomy, Perception and Cognition
Abstract: Reasons for mapping an unknown environment with autonomous robots are wide-ranging, but in practice, they are often overlooked when developing planning strategies. Rapid information gathering and comprehensive structural assessment of buildings have different requirements and therefore necessitate distinct methodologies. In this paper, we propose a novel modular Next-Best-View (NBV) planning framework for aerial robots that explicitly uses an explicit reconstruction quality objective to guide the exploration planning. In particular, our approach introduces new and efficient methods for view generation and selection of viewpoint candidates that are adaptive to the user-defined confidence objectives, exploiting the uncertainty encoded in a Truncated Signed Distance field (TSDF) representation of the environment. This results in informed and efficient exploration decisions tailored towards the predetermined objective. Finally, we validate our method via extensive simulations in realistic environments. We demonstrate that it successfully adjusts its behavior to the user goal while consistently outperforming conventional NBV strategies in terms of coverage, quality of the final 3D map and path efficiency.

Keywords: Path Planning, Navigation
Abstract: This paper introduces a novel framework for continuous 3D trajectory optimization (C-3TO) in cluttered environments, leveraging online neural Euclidean Signed Distance Fields (ESDFs). Unlike prior approaches that rely on discretized ESDF grids with interpolation, our method directly optimizes smooth trajectories represented by fifth-order polynomials over a continuous neural ESDF, ensuring precise gradient information throughout the entire trajectory. The framework integrates a two-stage nonlinear optimization pipeline that balances efficiency, safety and smoothness. Experimental results demonstrate that C-3TO produces collision-aware and dynamically feasible trajectories. Moreover, its flexibility in defining local window sizes and optimization parameters enables straightforward adaptation to diverse user’s needs without compromising performance. By combining continuous trajectory parameterization with a continuously updated neural ESDF, C-3TO establishes a robust and generalizable foundation for safe and efficient local replanning in aerial robotics.

Keywords: Path Planning, Autonomy, Control Architectures
Abstract: Real-time planning for fixed-wing unmanned aerial vehicles (UAVs) requires both rapid generation of dynamically feasible trajectories and the ability to reliably track those trajectories. Closed-loop rapidly-exploring random trees (CL-RRT) provide a promising framework by embedding vehicle dynamics and feedback control within rapid tree expansion; however, their application to fixed-wing aircraft remains limited. This work quantifies the impact of aircraft guidance and control architecture on both closed-loop node propagation and reference path tracking in real-time CL-RRT planning. Planning and tracking are first analyzed independently in simulation to isolate architectural effects. Results indicate that controller dynamics shape the short-horizon assessed reachable set during propagation, thereby constraining planner performance and computational robustness. Moreover, fixed-lookahead tracking assumptions break down under the nonuniform spacing and curvature of RRT waypoints. A Total Energy Control System (TECS) scheme and a modified Multi-Segment Adaptive Arc-Length Guidance (MS-AALG) law are proposed to address these shortcomings. Real-time CL-RRT simulations demonstrate improved reliability relative to conventional linear architectures.

Keywords: Perception and Cognition, Simulation, Sensor Fusion
Abstract: Monocular depth estimation suffers from inherent scale ambiguity, limiting its real-world applicability. While recent foundation models produce high-quality relative depth maps, their predictions lack metric consistency, which becomes critical in aerial environments. Recent rescaling approaches address this limitation by leveraging Digital Elevation Maps (DEM) as absolute references, yet existing DEM-based methods remain difficult to deploy due to their reliance on offline ground segmentation techniques and the absence of standardized validation frameworks. To overcome these challenges, we introduce a Synthetic-to-Geo-Real simulation framework that jointly models realistic flight dynamics and real georeferenced terrain, enabling a novel, systematic, and controlled evaluation of DEM-based rescaling across diverse operational conditions. The framework provides a unified benchmarking setting to compare multiple state-of-the-art depth models and alternative ground segmentation strategies under consistent assumptions. Our results show that semantic segmentation provides more consistent performance across models, altitudes, and environments, achieving up to a 60% reduction in error over geometric segmentation. We further observe that decoupling segmentation from depth estimation and selectively sampling ground points can improve scale recovery. These findings support the practical viability of DEM-based metric depth rescaling for Unmanned Aerial Navigation.

Keywords: Sensor Fusion, Perception and Cognition
Abstract: Autonomous drone flight within powerline corridors requires reliable onboard perception that can estimate and track the poses of conductors, even when individual cables temporarily leave the sensors’ field of view. In this work, we present an onboard perception system for mid-span corridor powerline pose estimation and tracking, building on previous research that combines mmWave radar and RGB camera measurements with flight controller odometry. First, we introduce a transformation of the cable direction that enables consistent estimation of the global powerline orientation regardless of the drone’s attitude by compensating for the mismatch between the camera and sensor planes. Second, we propose the Relative Cable Positions algorithm, which exploits the fixed geometric relationships between conductors to estimate the positions of cables outside the mmWave radar field of view based on measurements from cables that remain visible. The system is implemented onboard a drone and evaluated through real-world flight experiments conducted at a dedicated powerline test facility. The results show a clear reduction in position estimation error for conductors outside the radar field of view compared to odometry-only tracking under non-RTK conditions. Overall, the proposed methods improve the robustness of drone-based tracking and pose estimation of conductor geometry without relying on RTK, supporting safer and more reliable autonomous flight within mid-span powerline corridors.

Keywords: Perception and Cognition
Abstract: Visual Place Recognition is widely considered a mature field, yet most benchmarks are constrained to visually similar environments, predominantly urban or road scenes, leaving its performance in extreme environments largely unexplored. We address this gap by introducing a challenging benchmark built from Antarctic imagery, characterized by vast textureless areas covered by snow and ice, severe visual aliasing, and aerial perspectives from downward-facing drone cameras. We curate a GNSS-based ground truth refined with feature matching to establish a reliable evaluation. Additionally, we propose several adaptations to VPR methods to operate effectively under such conditions: rotation-robust VPR to accommodate aerial downward-facing viewpoints and a proxy to discard visually uninformative images. Finally, we present VPR-LAB, the most comprehensive pipeline to date for VPR benchmarking, enabling systematic evaluation across both conventional and unconventional datasets. Experiments demonstrate significant improvements for VPR in polar environments, directly enabling autonomous monitoring essential for Antarctic ecosystem conservation and multi-season ecological surveys.

Keywords: Autonomy, Perception and Cognition, Energy Efficient UAS
Abstract: The use of mobile robots is becoming increasingly common, especially for surveillance and search and rescue operations in disaster areas. During these operations, robots need to perceive the environment, detecting and estimating the pose of specific targets leveraging their on-board capabilities. Specifically, Unmanned Aerial Vehicles (UAVs), due to the limited payload, energy supply, and costs are often provided with low-cost and energy-efficient CPU-based companion computers, which may cause degraded performance, especially regarding vision applications. In this view, this work aims to present a CPU-based optimization of a real-time object position estimation suited for UAVs companion computers. The proposed pipeline, which implements quantization and inference optimization, has been tested over six state-of-the-art models for object detection on CPU-based hardware mounted on a drone. The validation has been carried out with real flights for estimating objects' positions.

Keywords: Perception and Cognition, Navigation, UAS Applications
Abstract: Direct visual odometry (DVO) is becoming widely used to estimate camera egomotion and simultaneously reconstruct/map the environment from image sequences, particularly due to its efficient use of photometric information and strong performance in weak-texture scenarios. However, since DVO relies directly on pixel intensity patterns for frame-to-frame alignment, it is inherently sensitive to brightness inconsistencies and illumination changes, which can degrade pose estimation, particularly in aerial robotics scenarios involving motion blur, illumination variations, and texture-poor environments. This paper proposes an adaptive pixel selection mechanism for DVO systems. The proposed approach prioritizes regions with strong structural content, ensuring that the pose estimate remains constrained by meaningful image gradients, even in low-texture or highly homogeneous scenarios. By integrating image texture descriptors, the system dynamically adjusts selection parameters, according to the scene complexity. This enables the operation across structured, semi-structured, and unstructured environments without introducing significant computational overhead. Experimental validation in both indoor and outdoor scenarios demonstrates that the proposed method preserves uniform spatial coverage while suppressing low-gradient and photometrically unstable regions. The results demonstrate enhanced trajectory consistency and greater robustness under challenging visual conditions, confirming the algorithm effectiveness for UAV navigation in GPS-denied environments.

Keywords: Perception and Cognition, Navigation, Sensor Fusion
Abstract: Inertial odometry is a compelling approach to state estimation for agile quadrotor flight due to its affordability, low weight, and robustness to perceptual degradation. However, relying only on integrated inertial measurements is impractical, as sensor errors and time-varying biases lead to significant pose drift. Although recent advances have enabled inertial odometry for drone racing, existing approaches show limited generalization to trajectories not seen during training. This work improves generalization by adopting a body-frame representation and incorporating body-frame torque dynamics into the learning pipeline. The learned model predicts short-horizon relative displacements, which are fused with IMU measurements in an Extended Kalman Filter. Experimental results show that the proposed method outperforms the previous state-of-the-art learned inertial odometry approach for quadrotor pose estimation on unseen trajectories.

Keywords: Perception and Cognition, Smart Sensors, UAS Applications
Abstract: The deployment of Unmanned Aerial Systems (UAS) in degraded visual environments relies heavily on Long-Wave Infrared (LWIR) sensors. However, thermal visual odometry is frequently compromised by high-frequency aerodynamic jitter and radiometric contrast loss. This paper presents a purely causal, real-time stabilization framework for 640x480 LWIR video at 30 fps, addressing the critical issue of "oversmoothing" inherent in standard Gaussian estimators. We identify a severe "Cold-Start Anomaly" in the classical Kalman Filter (KF), which, during turbulent initialization, suffers from numerical gain explosion and discards up to 16.31% of the useful frame area (Crop Ratio). To overcome this, we propose a spectrally- bounded H∞ minimax formulation that enforces strict numerical stability through an attenuator derived from the spectral radius of the observation-weighted information term. Furthermore, we validate a Covariance Persistence (Warm-Start) strategy that eliminates transient initialization penalties entirely. Evaluated across stable cruise, turbulent landing, and thermal crossover scenarios, the proposed H∞ framework demonstrated superior geometric preservation, restricting spatial loss to just 3.63% under cold-start turbulence, while matching the Kalman baseline in Inter-frame Transformation Fidelity (ITF). These results prove that spectrally-bounded robust filtering is essential for maintaining zero-latency, high-fidelity perception in autonomous UAS operations

Keywords: Autonomy, Path Planning, Multirotor Design and Control
Abstract: This paper proposes an autonomous flight system for small-sized drones using the efficient data structure based on circular buffered hash mechanism, and presents its flight experiment results. The proposed circular buffered hash data structure can be utilized in various algorithms handling voxel-like data, and it possesses the favorable characteristic of automatically maintaining a maximum number of the most recent data, thereby preventing a continuous increase in memory usage. The practical feasibility of the proposed data structure and the autonomous flight system has been verified by conducting flight experiments using a small-sized drone platform equipped with a Livox Mid-360 LiDAR, confirming the capability of achieving precise autonomous flight in GNSS-denied and cluttered environments.

Keywords: Energy Efficient UAS, Multirotor Design and Control
Abstract: Conventional multirotor aerial vehicles actively suppress yaw rotation during hover, expending power to maintain a fixed heading despite the fact that yaw regulation is not required for force balance or altitude control. This paper challenges that paradigm by proposing a spinning quadrotor architecture that intentionally operates at a sustained yaw rate, converting power traditionally spent on yaw regulation into useful aerodynamic effects. A dynamic model of the spinning quadrotor is developed, analysis for low Re range is conducted to choose an airfoil for lifting surfaces. Preliminary hardware tests show a 22% reduction in thrust required. These findings suggest that intentional yaw rotation, rather than being suppressed, can be exploited as a design mechanism for efficient and robust multirotor flight.

Keywords: Multirotor Design and Control, Aerial Robotic Manipulation, Control Architectures
Abstract: In robotics and biomechanics, trading metabolic cost for kinematic readiness is a well-established principle. This paper formalizes this concept for aerial multirotors through the introduction of aerodynamic promptness--a dynamic metric analogous to dynamic manipulability in robotics. By formulating redundancy resolution as a geometric multi-objective optimization along task fibers, we rigorously characterize the topological trade-off between energy consumption and promptness. We demonstrate that this interplay is fundamentally governed by fiber geometry. Cooperative actuation regime yields compact fibers with bounded, compatible Pareto fronts. Conversely, antagonistic actuation regime unlocks unbounded fibers, enabling aerodynamic co-contraction that drives promptness to hardware limits at the expense of flight endurance. We establish a structural isomorphism between aerodynamic co-contraction and biologically inspired variable stiffness actuators, introducing a dynamic ``flying muscle'' paradigm. Ultimately, this framework transitions multirotor allocation from heuristic energy minimization to principled, geometry-aware Pareto navigation, laying foundational theory for the design and control of highly agile aerial platforms.

Keywords: Path Planning, Perception and Cognition
Abstract: Traditional coverage path planning methods for unmanned aerial vehicles (UAVs) take an overly simplistic look at the sensor footprint of the camera, resulting in inefficient path plans that waste significant coverage and are not optimal in travel time. We propose a coverage path planning algorithm based on generating and sequencing a set of stationary vantage points that the UAV will travel to and capture imagery. These vantage points ensure that the coverage path plan achieves a spatial resolution threshold throughout the entire search domain while minimizing the amount of unnecessary excess coverage both inside and outside of the search domain. We also propose a routing method for maximizing the portion of the search domain we can cover when faced with a maximum mission time constraint. Simulation testing shows the improvement of our method in both coverage efficiency and total travel time compared to lawnmower-based coverage path plans. Experimental testing details how coverage path plans based on stationary coverage can degrade when the UAV's onboard camera does not have a level attitude.

Keywords: Security, Reliability of UAS, UAS Applications
Abstract: Given the increasing use of Unmanned Aerial Vehicles (UAVs), cyberattacks targeting them have caused significant financial and operational losses, highlighting the need for effective security solutions. Despite advances in Machine Learning (ML) approaches, the literature still lacks investigations into the generalizability of results across datasets for detecting Global Navigation Satellite Systems (GNSS) Spoofing. The main contribution of this study is to show that different collection conditions, experiments, receptors, and even the collection location itself can prevent an ML-trained model from generalizing to another dataset. On the TEXBAT and OAKBAT datasets, we propose a Transfer Learning (TL) approach that uses a model trained on TEXBAT and fine-tuned on OAKBAT, demonstrating strong cross-dataset generalization. In contrast, when evaluated without fine-tuning (FT), we observed a substantial performance drop, highlighting a fundamental gap in approaches that rely on single-dataset evaluation or on independent models for each dataset. Such practices limit real-world applicability, particularly for UAV systems operating in heterogeneous and complex environments. Our approach was tested across six distinct OAKBAT scenarios, achieving accuracy above 99% in four of the six cases, indicating potential for practical deployment in real-world UAV applications.

Keywords: UAS Applications
Abstract: Autonomous UAVs enable safe inspection of hazardous underground environments, but simultaneous GNSS denial and visibility degradation from airborne dust present critical navigation challenges. Conventional visual odometry fails under feature scarcity, while LiDAR-based methods suffer from noise artifacts as suspended particles are misinterpreted as obstacles. This paper presents a UAV system for autonomous mapping in dust-laden, GNSS-Denied tunnels. LiDAR-Inertial odometry was integrated with IMU pre-integration to achieve 100~Hz state estimation for responsive control in confined spaces. An intensity-based dust filtering algorithm removes point cloud contamination in real-time, enabling autonomous capabilities despite visibility degradation. Experimental validation in a real mining tunnel under controlled visually degraded conditions demonstrates autonomous operation. The system completes exploration and mapping in 480 seconds, achieving 0.19 m mean mapping accuracy and 0.60 m RMS error against Total Station ground truth. Results establish quantitative benchmarks for UAV deployment in underground construction and emergency response scenarios where human access is restricted.

Keywords: Multirotor Design and Control, Micro- and Mini- UAS, Control Architectures
Abstract: Adaptive dynamic programming (ADP) and policy iteration (PI) algorithms are powerful tools for data-driven optimal control. When using the linear-quadratic regulator (LQR) in the ADP framework, most existing approaches assume disturbance-free system dynamics. Data-driven design eliminates the need for a dynamical model by utilizing data from the system state and input. If the system is disturbed by an unknown disturbance, the system input is not fully known, which makes standard ADP and PI approaches impractical. In this study, a novel PI-based ADP framework is proposed to explicitly handle constant unknown disturbances by integrating frequency-domain filtering with the classical ADP approach. High-pass filtering of input-output trajectories suppresses the effects of disturbances on the design, allowing the use of established PI techniques, while a low-pass filter extracts the required steady-state feed-forward term. The resulting control law preserves the LQR structure but improves steady-state accuracy and robustness. The process is experimentally validated using a quadcopter altitude control design with unknown mass and motor dynamics. The unknown gravitational force is assumed to be a constant unknown disturbance input. The results show that the proposed approach eliminates steady-state altitude errors compared to baseline controller gains. These findings demonstrate that disturbance-aware ADP provides a practical and effective framework for robust data-driven control in systems with unknown dynamics.

Keywords: Multirotor Design and Control, Control Architectures, UAS Applications
Abstract: This paper presents a robust adaptive attitude control framework for quadcopters subject to parametric uncertainty and unmodeled nonlinear dynamics. The control design is based on a backstepping approach, where the uncertain moments of inertia and the synaptic weights of a radial basis function neural network (RBFNN) that is employed to approximate the unknown dynamics, are estimated online through dedicated adaptive laws. Projection operators are incorporated to ensure strict positivity of the inertia estimates, thereby preserving physical consistency and controllability, and ultimate boundedness of the RBFNN weights. Moreover, through rigorous analysis, uniform ultimate boundedness of the tracking errors and boundedness of all closed-loop signals is formally established. The proposed controller is experimentally validated on a real-world quadcopter platform directly utilizing measurements from onboard sensors, in contrast to the vast majority of the existing literature that relies on high-precision motion capture systems. Finally, comparison results against alternative controllers validate the superiority of the proposed control protocol.

Keywords: Multirotor Design and Control, Control Architectures
Abstract: Recently, Reinforcement Learning (RL) has been applied to numerous robotics domains, including end-to-end quadrotor control. The training of useful policies using RL usually involves the time-consuming and often unprincipled, tuning-based design of a reward function. In this work, we present a novel alternative method to train an end-to-end quadrotor control policy using a constant reward function r(s,a)=1. Since a constant reward function cannot carry information, we show that end-to-end policies can learn to fly solely through the feedback of the termination signal. Furthermore, we show that encoding additional objectives into the termination signal leads to robust end-to-end policies with low-level RPM outputs. We demonstrate that these policies can be directly transferred to a real quadrotor and even generalize to new tasks, such as trajectory tracking. Finally, we compare the trajectory-tracking performance of our policy to other classical and RL-based methods and find that our policies can achieve similar performance to other RL-based approaches while eliminating the need for hand-tuned reward functions. We open-source our implementation for the benefit of the community and to advance research in this area.

Keywords: Multirotor Design and Control, Path Planning
Abstract: This paper presents an aggressiveness-aware control framework for quadrotor UAVs that integrates learning-based oracles to mitigate the effects of unknown disturbances. Starting from a nominal tracking controller on mathrm{SE}(3), unmodeled generalized forces and moments are estimated using a learning-based oracle and compensated in the control inputs. An aggressiveness-aware gain scheduling mechanism adapts the feedback gains based on probabilistic model-error bounds, enabling reduced feedback-induced aggressiveness while guaranteeing a prescribed practical exponential tracking performance. The proposed approach makes explicit the trade-off between model accuracy, robustness, and control aggressiveness, and provides a principled way to exploit learning for safer and less aggressive quadrotor maneuvers.

Keywords: Multirotor Design and Control, UAS Applications, Micro- and Mini- UAS
Abstract: Reference accelerations are tracked by the action of the inner-loop attitude control of quadrotors. While conventional PID controllers are widely utilized, they often lack the robustness required to handle modeling mismatches and aggressive maneuvers. This work proposes a cascaded control architecture where the inner-loop attitude rate control is governed by a Super-Twisting Sliding Mode Controller (ST-SMC). This high-order sliding mode control strategy enhances robustness while mitigating the chattering typically associated with first-order sliding mode controllers. Experimental validation on a Crazyflie platform under stress scenarios, including carrying an unmodeled payload from an offset from the center of mass and navigating with damaged propellers, demonstrates that ST-SMC significantly outperforms the industry-standard PID, achieving a 15.58% reduction in position-tracking RMSE and a 22.93% reduction in velocity-tracking error. Furthermore, the ST-SMC demonstrated reduced control effort and superior reliability, leading to fewer crashes during agile flights.

Keywords: Multirotor Design and Control, UAS Applications, Micro- and Mini- UAS
Abstract: Linearization around the near-hover condition is a common approach in modeling quadrotors, assuming small pitch and roll angles to yield a fully actuated dynamic model. Such linearized models are particularly useful because they can be readily applied to control off-the-shelf quadrotors, enabling their use in commercial applications and academic research. For the control of translational motion and heading angle, most models reported in the literature require eight parameters to be identified via a black-box procedure. In this paper, we revisit a recently proposed simplified model that requires identifying only four parameters while still achieving good performance. Based on this model structure, a model reference adaptive controller is proposed to handle parameter variations and improve tracking performance during more aggressive navigation. Experimental results obtained with a quadrotor platform validate the proposed approach by comparing the trajectory-tracking performance of the model requiring the identification of four parameters against that of the model requiring the identification of eight parameters and a purely analytical model. The obtained results show that the proposed model is not only simpler and easier to implement but also capable of delivering superior performance, thus providing an interesting framework for researchers working with off-the-shelf quadrotors.

Keywords: Path Planning, Regulations, Integration
Abstract: We study routing of Urban Air Mobility (UAM) flights while ensuring robustness to large-scale disruptions during which many aircraft may need to initiate contingency maneuvers at the same time. We propose algorithmic solutions for routing drones so that they always stay sufficiently close to potential alternate landing locations, while preventing the situations when too many drones rely on a single landing site at once. The performance of our algorithms is evaluated on simulated urban scenarios.

Keywords: Path Planning, Autonomy, Multirotor Design and Control
Abstract: Autonomous unmanned aerial vehicles (UAVs) increasingly operate in cluttered environments where global planners such as RRT* are not directly deployable at control rates. This paper presents a real-time local planning and obstacle avoidance module for an omnidirectional multirotor (omnicopter) by extending the Dynamic Window Approach to six degrees of freedom (6D-DWA). Our method achieves real- time feasibility through (i) local-map voxelisation, (ii) a compact sphere-based approximation of the vehicle geometry, and (iii) adaptive velocity sampling in the 6D search space. To improve reactivity to unknown obstacles, we introduce a context-aware “Agile Mode” that adjusts scoring weights online to trade- off between goal progress, clearance, and heading/facing con- straints during evasive manoeuvres. We evaluate our approach in simulation across computational stress tests, dense-waypoint path tracking, and static/unknown obstacle scenarios. Our planner runs consistently within a 0.2 s control loop, tracks waypoint-dense global paths with < 0.1 m average cross-track error and ∼ 13◦ average heading error, and avoids collisions in static environments. For unknown obstacle avoidance, Agile Mode achieves 79.3% success for an off-centre obstacle and 41.4% for a centred obstacle, highlighting both the effectiveness of adaptive weighting and remaining limitations in highly constrained geometries.

Keywords: Path Planning, UAS Applications, Environmental Issues
Abstract: This work addresses the seed dispersal planning problem as a Commodity-Constrained Split Delivery Vehicle Routing Problem (C-SDVRP), in which the UAV must repeatedly return to a depot for reloading and recharging. We benchmark five routing algorithms across field sizes from 25 × 25 m to 150 × 150 m: Nearest-Neighbor (NN), two variants of Lin- Kernighan with split (LKH-Split), Discrete Artificial Hummingbird Algorithm (D-AHA), and Hybrid Genetic Search (HGS). Results highlight trade-offs between computational efficiency,solution quality, and scalability. LKH-Split achieves sub-second planning times but incurs relatively high trajectory costs, withoptimality gaps ranging from 41.6% to 47.2%. D-AHA shows competitive performance on small-to-medium instances, achieving low optimality gaps (1.4%–6.2%), but its performance degrades significantly in large fields (46.9%–57.0% optimality gap). Notably, the simple NN heuristic outperforms more complex metaheuristics in the largest instance (150 × 150 m), reachinga near-optimal solution with a 0.2% optimality gap and sub-millisecond computation time. These results provide practical guidelines for selecting routing algorithms based on field scale and mission constraints in UAV-based aerial seeding operations.saw

Keywords: Path Planning, Security, Risk Analysis
Abstract: The defense of civil and military airspace is increasingly challenged by low-cost, small Unmanned Aerial Vehicles (UAVs) with high maneuverability, enabling them to evade modern detection systems, disrupting the airspace, or even leading to a catastrophe. Contrary to existing counter UAS (cUAS) methods that focus on optimal sensor placement or control, this paper identifies vulnerabilities in existing sensor deployments by formulating the problem as a zero-sum dif- ferential surveillance–evasion game between an intruding UAV and the detection system. In this formulation, intruders seek paths that minimize detection probability while reaching their goal. Our focus is on identifying optimal intruder best response trajectories given the knowledge of the detection system to evaluate its performance and vulnerabilities. However, planning such trajectories is challenging due to the high dimensionality of the space–time domain and the non-convex visibility regions of panning sensors. To address this, we propose a hybrid approach that combines sampling-based planning with a non- linear program (NLP) to obtain a locally optimal solution. The sampling solution provides an initial guess, thereby alleviating the computational burden of the NLP and refining the solution to obtain locally optimal paths. Finally, we use a Monte Carlo simulation to demonstrate that our proposed attacker’s best response approach reduces the detection probability by 49.6% on average relative to purely sampling-based methods, with an additional average cost of 53.8 seconds per trial.

Keywords: Path Planning, Autonomy, Navigation
Abstract: This paper presents a UAV path planning framework designed to enable smooth, collision-free autonomous navigation in static, GPS-denied environments. The framework relies on point cloud data collected by a primary, pilot UAV, which uses onboard sensing, SLAM, and object detection to incrementally map an unknown environment and identify waypoints. Path planning is performed in a bounded SE(3) state space using sampling-based algorithms with a multi-objective formulation that balances path length, obstacle clearance, and orientation smoothness. Safety is ensured through tight integration with a dedicated collision detection and proximity query engine, enabling efficient collision and distance queries against the octree-based environment representation. Planned paths are further refined through shortcutting and B-spline smoothing to produce dynamically feasible trajectories, which are executed via a MAVROS–PX4 control pipeline. The enhanced collision validation approach for global path planning incorporates 6-DOF vehicle motion to facilitate accurate navigation. Simulation and flight test results demonstrate the effectiveness of the proposed framework for reliable UAV autonomous navigation in complex and GPS-denied environments.

Keywords: Path Planning, Risk Analysis, Simulation
Abstract: Route planning for Unmanned Aerial Vehicles (UAVs) in hostile surveillance environments requires trajectory optimization frameworks that incorporate realistic radar detection physics beyond simple geometric avoidance. This study proposes a novel route planning framework for bistatic and multistatic radar environments using Segmented Particle Swarm Optimization (PSO). Unlike traditional approaches that rely on monostatic radar models, we formulate a high-fidelity observability metric based on bistatic range-rate sum physics and continuous Doppler-notch attenuation. This formulation enables the planner to exploit ”blind speeds” and clutter rejection filters inherent in Pulse-Doppler radars. We demonstrate the effectiveness of the proposed method across four scenarios, ranging from canonical iso-range validation to complex multistatic environments with strict time-of-arrival constraints. The results show that the planner autonomously discovers low-observability maneuvers, such as iso-range loitering and bistatic weaving with low RCS, significantly reducing the tracking probability while satisfying kinematic limits and mission timing requirements.

Keywords: Perception and Cognition, Sensor Fusion, Training
Abstract: Multi-UAV collaborative perception is essential for maintaining safe operations in complex urban environments, where single-agent vision often suffers from instability and the resulting localization drift in adverse weather conditions. This paper presents a Res-Gated Fusion framework that integrates residual structures with a dynamic gating mechanism to dynamically prioritize reliable geometric attributes when visual cues are degraded, thereby enhancing perception robustness. Building upon this, we introduce a risk representation that transforms geometric predictions into conflict indicators for collaborative decision support. Extensive evaluations on the customized CityUAV-3D benchmark demonstrate that multi-UAV collaborative perception consistently outperforms individual sensing in both perception accuracy and conflict detection by leveraging multi-view complementarity. Furthermore, strong zero-shot cross-scene transfer results in unseen complex scenarios underscores the framework's robustness and its potential for deployment in diverse urban airspace.

Keywords: Perception and Cognition, UAS Communications, Security
Abstract: Unmanned Aerial Vehicles (UAVs) increasingly rely on wireless communication links for command, control, and data exchange, making them vulnerable to cyber-attacks that may compromise operational safety. This paper presents a learning-based perception framework for detecting anomalous UAV communication behavior using the UAV-NIDD dataset. The proposed approach treats network traffic analysis as a cyber-perception problem, where machine learning models infer deviations from normal communication patterns in real time, and includes a mitigation strategy based on explainable AI to define the suitable mitigation actions. Supervised algorithms including Random Forest, XGBoost, Support Vector Machine, and Logistic Regression are evaluated for both binary (normal vs. attack) and multi-class (attack types) intrusion detection. Experimental results demonstrate near-perfect performance in binary classification and high macro-F1 scores in multi-class scenarios, highlighting the effectiveness of tree-based ensemble models in capturing non-linear packet-level patterns. Feature importance analysis and SHAP-based explainability reveal that transport-layer attributes (e.g., UDP packet length and source port), together with wireless-layer indicators, provide strong discriminative signals for identifying malicious activity. The findings show that UAV cyber-attacks exhibit structured statistical signatures that can be effectively perceived through data-driven learning models. This work contributes toward enhancing cyber-situational awareness and strengthening safety assurance in increasingly autonomous aerial operations.

Keywords: Perception and Cognition, Micro- and Mini- UAS, Manned/Unmanned Aviation
Abstract: The detection of concealed drones in complex environments, such as urban areas or forests, remains a significant challenge for computer vision systems. While the Real-Time Detection Transformer (RT-DETR) achieves state-of-the-art performance in general object detection, its reliance on global self-attention may limit its effectiveness for targets that blend into the background. This work investigates the impact of integrating alternative attention mechanisms into the RT-DETR architecture to enhance the detection of concealed drones. We evaluate four distinct attention modules—Multi-Head Self-Attention (MHSA), Convolutional Block Attention Module (CBAM), Window Attention (WA), and Local-Global Attention (LGA) - prior to the encoder, as well as Efficient Channel Attention (ECA) prior to the decoder. Using a test set biased towards camouflaged drones, we demonstrate that the placement and type of attention are critical to performance. Specifically, applying Window Attention to lower-level feature maps improves Average Precision (AP) by 1.5%, while the implementation of ECA at the encoder output yields a gain of 2.0% in AP. These findings suggest that for concealed drone detection, local and channel-selective attention mechanisms are superior to global self-attention, provided they are applied at semantically rich feature levels.

Keywords: Perception and Cognition, Networked Swarms, Control Architectures
Abstract: This paper presents the experimental implementation of a gesture-based formation control framework for a multi-UAV system using a Natural User Interface (NUI). Hand gestures are recognized through a Multilayer Perceptron (MLP) trained on skeletal hand features extracted with MediaPipe Hands, enabling intuitive high-level command generation. The recognized gestures define both formation configurations and motion references for a leader–follower architecture, while low-level control is achieved through input–output linearization and consensus-based formation strategies. Experimental validation with three Crazyflie quadrotors and an OptiTrack motion capture system demonstrates stable formation transitions and coordinated collective motion generated through natural human interaction. Results show convergence toward desired formations and successful trajectory tracking under dynamic gesture commands, highlighting the feasibility of integrating vision-based human interaction with cooperative aerial robotics.

Keywords: Perception and Cognition, Navigation, UAS Applications
Abstract: Plane structures such as floors, ravines, and surfaces play a fundamental role in robotic navigation, perception, and mapping. Although single-frame plane segmentation is well established, ensuring temporal consistency and geometric stability across sequential observations remains a challenging problem. In this work, we propose a stability-driven pipeline for multi-plane segmentation, temporal tracking, and semantic classification in 3D point clouds. The method first extracts dominant planes using iterative RANSAC, followed by a geometry-aware association strategy that tracks planes over time based on orientation, centroid, and offset constraints. To evaluate temporal robustness, we introduce a comprehensive stability analysis framework that measures global angular deviation, frame-to-frame variation, metric offset evolution, and inlier support over time. Experiments on real-world irregular terrain demonstrate that the proposed approach maintains consistent plane orientation and parameterization despite altitude and viewpoint changes or partial occlusions. The results show bounded angular drift, smooth metric evolution, and sustained geometric support, confirming the robustness of the proposed temporal modeling strategy.

Keywords: Autonomy, Perception and Cognition, Control Architectures
Abstract: Navigating UAVs in environments like tunnels or mines is a complex task due to unavailable GNSS for localization, uneven or absent lighting, and likely scarce wall features, especially at high speeds. In this paper we propose a novel proof-of-concept reactive UAV navigation technique using only LiDAR data that combines geometric and machine-learning algorithms.

2D range information is processed by a deep neural network to establish the UAV's yaw relative to the tunnel's longitudinal axis for navigation directions. Additionally, a geometric method computes the safest location inside the tunnel that maximizes distance to the closest obstacle. This information proves to be sufficient for simple yet effective navigation in straight and curved environments at speeds of up to 6,m/s in simulation.

Keywords: Control Architectures, Navigation
Abstract: In this paper, we study a navigation problem composed of two distinctive phases required to perform a task. Navigation in the first phase and the decision to transition to the second phase directly influence the chance of success of the second phase and, consequently, the entire task. Specifically, we consider an intercept task involving two Dubins vehicles and two target sets. We demonstrate that the problem can be solved by a set of interdependent stochastic Hamilton-Jacobi-Bellman and Backward Kolmogorov equations. To enable the use of these equations, we exploit an exponential utility function and Cantelli’s inequality for probabilities. Finally, we compute the solution and validate it through numerical simulations.

Keywords: Control Architectures, Autonomy, Path Planning
Abstract: This paper presents a systematic framework for computing formally guaranteed trajectory tracking error bounds for autonomous helicopters based on Robust Positive Invariant (RPI) sets. The approach establishes a closed-loop translational error dynamics which is cast into polytopic linear parameter-varying form with bounded additive and state-dependent disturbances. Ellipsoidal RPI sets are computed, yielding explicit position error bounds suitable as certified buffer zones in upper-level trajectory planning. Three controller architectures are compared with respect to the conservatism of their error bounds and tracking performance. Simulation results on a nonlinear helicopter model demonstrate that all architectures satisfy the derived bounds, while highlighting trade-offs between performance and the conservatism of the computed invariant set.

Keywords: Control Architectures, Multirotor Design and Control, Simulation
Abstract: From a maneuverability perspective, the main advantage of tilting multirotor UAVs lies in the dynamic variability of the feasible executable wrench, which represents a key asset for physical interaction tasks. Accordingly, cant-angle selection should be optimized to ensure high performance while avoiding abrupt variations and preserving real-world feasibility. In this context, this work proposes a lightweight control framework for star-shaped interdependent cant-tilting hexarotor UAVs performing interaction tasks. The method uses an offline-computed look-up table of zero-moment force polytopes to identify feasible cant angles for a desired control force and select the optimal one by balancing efficiency and smoothness. The framework is integrated with a geometric full-pose controller and validated through Monte Carlo simulations in MATLAB/Simulink and compared against a baseline strategy. The results show a significant reduction in computation time, together with improved pose-tracking performance and competitive actuation efficiency. A final physics-based simulation of a complete wall inspection task in Simscape further confirms the feasibility of the proposed strategy in interacting scenarios.

Keywords: Control Architectures, Multirotor Design and Control, Simulation
Abstract: Commercial quadrotors typically expose to the pilot a reduced set of control inputs that correspond to intuitive first-order motion commands, such as the desired translational velocity expressed in a heading-aligned frame and the desired yaw rate. The underlying onboard autopilot is responsible for mapping these pilot commands into thrust and moment references that command the full nonlinear quadrotor dynamics. Motivated by this industrial practice, this paper formulates the problem of designing an onboard controller that emulates the closed-loop behavior of commercial velocity-commanded quadrotors. A pilot-level kinematic reference model is introduced, in which the commanded heading frame directly drives the translational position, yaw dynamics, velocity, and yaw rate. Using the standard rigid-body quadrotor model, it is shown, via Lyapunov stability analysis, that this behavior can be emulated by regulating a velocity-tracking error defined in the heading frame. This formulation naturally leads to a cascaded control architecture, consistent with commercial autopilot implementations, and provides a principled framework for reproducing commercial flight behavior on custom quadrotor platforms. Software-in-the-loop (SIL) simulations with X-Plane providing the flight dynamics model and MATLAB/Simulink executing the control algorithm, along with experimental results, validate the proposed controller.

Keywords: Control Architectures, Navigation, UAS Applications
Abstract: This work addresses the use of the null-space-based behavioral control technique to design a controller for guiding a single quadrotor during the transport of a cable-suspended load. The adopted control paradigm is the virtual structure one, for which the virtual structure is the imaginary vertical line connecting the drone and the load. The control objective is to maintain the virtual structure in the vertical position during task execution, thereby minimizing load swing. The designed controller is discussed in detail, and experimental results from a quadrotor equipped with this controller to transport a cable-suspended load in an indoor environment are presented and analyzed, thereby validating the approach adopted for the application. To obtain the necessary feedback, namely the position and velocity of the quadrotor and the load, a motion capture system is used, enabling

Keywords: Control Architectures, UAS Applications, Autonomy
Abstract: Reliable and safe trajectory tracking control is critical for many missions involving fixed-wing unmanned aerial vehicles (FW-UAVs). This paper presents a simple, safe, and practical solution to the trajectory tracking problem for FW-UAVs, where the vehicle is required to track a time-parameterized curve defined by a sequence of waypoints. By leveraging the built-in path-following capabilities already available in widely used autopilot systems such as PX4, the proposed approach reduces the trajectory tracking problem to speed command regulation. Instead of redesigning low-level controllers, the method achieves time coordination along the trajectory by appropriately commanding the vehicle’s airspeed, while relying on the autopilot’s internal guidance and stabilization loops for lateral and altitude tracking. The proposed solution is validated through extensive Monte Carlo simulations in the Gazebo simulator, as well as real flight experiments with both single and multiple (Vertical Take-Off and Landing) VTOL UAVs. The experimental scenarios include missions requiring precise trajectory tracking to achieve simultaneous destination arrival, both along the path and at the final waypoint. Simulation and flight test results demonstrate that the proposed method provides reliable, safe, and accurate trajectory tracking performance.

Keywords: Micro- and Mini- UAS, Navigation, Autonomy
Abstract: Micro-drones under 250g offer unique potential for exploring confined subterranean environments such as mines, tunnels, and caves, where larger robots cannot operate. However, complete autonomous explore-and-return navigation in such GPS-denied spaces remains highly challenging for these platforms: existing topological and metric SLAM methods require lidar-class sensors or GPU-class compute, exceeding strict size--weight--power budgets. We present a novel topological navigation pipeline designed from the ground up for sparse depth sensing and CPU-only compute. Our approach builds confidence-aware radial profiles from sparse ToF arrays, applies potential-based stabilization to filter transient junction detections, and adapts landmark matching to accumulated odometry drift through covariance-scaled search gates. Rotation-invariant Fourier descriptors enable re-identification during return navigation despite different approach headings. We validate the system using hardware-in-the-loop simulation on the target embedded platform (RADXA Zero 2 Pro, ARM Cortex-A55), achieving 2,Hz perception cycles. To our knowledge, this is the first complete explore-and-return pipeline for sub-250g platforms using sparse ToF-based event perception and lightweight onboard odometry.

Keywords: Micro- and Mini- UAS, UAS Applications, UAS Testbeds
Abstract: Virtual Environments and Extended Reality (XR) have transformed artificial intelligence and robotics research by enabling realistic simulations for training and testing. XR enables the seamless integration of digital and physical spaces, offering new opportunities for training and testing autonomous systems. In this work, we propose RoboticsXR, an XR-based framework where a robot operates in a physical lab while images are captured from a virtual environment. This approach fuses real-world motion with synthetic visual data to develop and validate visual navigation algorithms efficiently. RoboticsXR feasibility is assessed using a convolutional neural network trained on synthetic images from NVIDIA Isaac. Performance is compared across different hardware platforms, and using virtual and real camera images. Real-time performance is evaluated for a UAV moving in a confined test area, while virtual images from a synthetic mountain environment are shared with the robotic platform. Results highlight RoboticsXR's potential to extend laboratory testing capabilities, reduce costs, and improve the reliability of vision-based navigation in autonomous robotics systems.

Keywords: Micro- and Mini- UAS, Autonomy, Perception and Cognition
Abstract: Autonomous landing in cluttered or unstructured environments remains a safety-critical challenge for unmanned aerial vehicles (UAVs), particularly under noisy perception caused by sensor uncertainty and platform-induced disturbances such as vibration. This paper presents an evidence-based probabilistic framework for autonomous UAV landing that explicitly separates decision-making under uncertainty from execution via visual servoing. Landing safety is modeled as a latent variable and inferred through recursive accumulation of frame-wise visual likelihoods derived from flatness, slope, and obstacle cues, yielding a temporally consistent belief map that is robust to transient perception errors. Physical feasibility is enforced through a hard geometric constraint based on the minimum required landing radius of the UAV, ensuring that undersized but visually appealing regions are rejected. The final landing site is selected using constrained maximum a posteriori estimation. Once selected, the UAV locks onto the target region using ORB feature tracking and performs precise alignment and descent via image-based visual servoing (IBVS). The proposed approach is validated through both real-world laboratory experiments and high-fidelity simulations in Nvidia Isaac Sim, demonstrating consistent, cautious, and stable landing behavior across domains.

Keywords: Micro- and Mini- UAS, Swarms
Abstract: This paper addresses distributed formation control for multiple aerial robots. We design a new distance-based control strategy to steer a team of UAVs toward equally spaced planar configurations, with particular focus on regular polygon-shaped formations. The distributed controller, based on attractive-repulsive potential fields, does not require prescribed desired distances and enables real-time formation reconfiguration when the team composition changes, preserving equal spacing. The Lyapunov/LaSalle framework is used to analyze the stability of the proposed control, and equilibrium configurations are investigated via Hessian-based analysis. Real-world experiments with a team of UAVs validate the efficacy of the proposed gradient-based control and demonstrate convergence to stable reconfigurable polygon formations.

Keywords: Biologically Inspired UAS
Abstract: Flapping-Wing Micro Aerial Vehicles (FWMAVs) 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 inability to carry heavy payloads. The unsteady aerodynamics governing these vehicles also significantly determine their performance, driven by their flapping motion and flexible wing membrane. This makes the material selection for the wing membrane a key aspect in the design of these vehicles. However, a systematic comparative analysis of membrane material properties remains a gap in the literature, especially for larger bird-like FWMAVs. This study presents an experimental investigation of the aerodynamic characteristics of an FWMAV with wings made from different flexible materials. Systematic testing has been conducted using a load cell to evaluate their performance and the influence of the wing membrane material. Wings were tested for lift and thrust generation across a range of flapping frequencies and also at a constant 25% throttle setting. Mylar (PET) wing yielded the highest averaged lift over one flapping cycle and demonstrated a lift-to-weight ratio of 0.747 at a flapping frequency of 3.75 Hz, outperforming Ripstop Nylon by a factor of 4.7. Whereas Nylon showed better thrust performance, especially at higher flapping frequencies, producing 1.170~N net thrust at 3.75 Hz, nearly 180% higher than the 0.420 recorded for Ripstop Nylon. Results indicate that the wing membrane significantly influences the vehicle's performance, and that a systematic selection is crucial given the mission requirements. This work provides insight into the wing membrane effect on performance and enhances current knowledge of FWMAV wing design.

Keywords: Biologically Inspired UAS, Micro- and Mini- UAS
Abstract: The conceptual design of Flapping-Wing Uncrewed Aerial Vehicles (FWUAVs) is currently constrained by the trade-off between computational cost and modeling fidelity. While high-fidelity solvers like Unsteady Vortex Lattice Methods (UVLM) capture the complex physics of flapping flight, they are too computationally expensive for iterative design optimization. To address this gap, this paper introduces a novel automated inverse-design methodology that directly translates high-level mission requirements into optimized physical hardware specifications. We propose a bi-level optimization framework coupled with a Gaussian Process Regression (GPR) aerodynamic surrogate model, enabling instantaneous force predictions across the design space. The architecture's outer level performs a global search over wing geometry, while the inner level solves for trim-feasible flight kinematics to ensure every candidate design is physically viable. To ensure the reliability of the proposed designs, the underlying aerodynamic and power models are validated against experimental data from three ornithopter prototypes, achieving endurance predictions with a low margin of error relative to the observed values across scales.

Keywords: Autonomy, Navigation, Control Architectures
Abstract: In aerial robotics, data-driven Reinforcement Learning (RL) approaches have proven highly effective for obstacle avoidance and goal-directed navigation, especially when operating on high-dimensional sensor data that provide only partial, local information about the environment. Such limited observability, combined with irregularly shaped obstacles, poses significant challenges for reactive control policies that rely solely on instantaneous observations. To address these issues, this paper introduces a Twin Deep Deterministic Policy Gradient (TD3)-based algorithm that leverages explicit Temporal Augmentation of the Observation space (TAO-TD3). The proposed method preserves the simplicity of the original TD3 framework by augmenting the observation with a short history of past states and incorporating a lightweight recurrent network, without requiring changes to the TD3 training paradigm. Extensive simulations across diverse environmental topographies and irregular obstacle shapes demonstrate that the proposed approach nearly halves the collision rate and improves overall navigation success compared to feedforward RL-based architectures.

Keywords: Navigation, Path Planning, Autonomy
Abstract: Autonomous navigation of multi-rotor unmanned aerial vehicles (UAVs) in confined and cluttered three-dimensional (3D) environments requires balancing goal reaching efficiency with safety constraints related to obstacle avoidance and workspace boundaries. Conventional navigation methods can be limited by model dependence, replanning cost, and reduced adaptability when obstacle configurations vary. Deep reinforcement learning (DRL) provides a flexible closed-loop decision-making framework, but standard reward-based methods do not explicitly constrain safety violations. To address this issue, this paper proposes Conservative Cost-Aware Soft Actor-Critic Lagrangian (CCSAC-Lag), a safe DRL framework that combines Soft Actor-Critic (SAC) with a Lagrangian cost-constraint mechanism. The navigation task is formulated as a constrained Markov decision process (CMDP), where safety violations are represented by an independent indicator cost rather than being merged into the reward. To reduce cost underestimation under sparse safety events, CCSAC-Lag employs twin cost critics with a conservative maximum backup over cost estimates. The proposed method is evaluated in a physics-based ROS--Gazebo simulation environment with randomized obstacle configurations. Experimental results show that CCSAC-Lag reduces safety violations and improves the success rate compared with standard SAC, while maintaining reasonable navigation efficiency. These results suggest that conservative cost-aware learning can improve safety compliance for multi-rotor navigation in bounded 3D environments.

Keywords: Navigation, Technology Challenges, Simulation
Abstract: This work presents a lightweight, high-level planner for UAV navigation based on a Soft Actor-Critic (SAC) reinforcement learning policy. Trained in a simplified 3D simulation, the policy generates velocity commands to reach static goals and generalizes zero-shot to dynamic tasks, including trajectory tracking and pursuit of moving targets. A parallelized PyTorch implementation accelerates training, enabling convergence in under five minutes on accessible computing hardware. The policy was validated in SITL and controlled indoor flight experiments using a real UAV with Vicon-based localization. Results demonstrate that a policy trained under simplified assumptions can generalize to multiple navigation-related tasks while requiring modest onboard computational resources. A video demonstration of the main experiments can be found at https://youtu.be/R2PkxlgmO74.

Keywords: Navigation, Sensor Fusion, Integration
Abstract: Reliable navigation in Global Navigation Satellite System (GNSS)-denied environments remains a critical challenge for fixed-wing Unmanned Aerial Vehicles (UAVs). In such conditions, Inertial Navigation Systems (INSs) must be aided by heterogeneous sensors to mitigate drift and ensure long-term accuracy. This paper presents the development, implementation, and experimental validation of a unified Error-State Extended Kalman Filter (ESEKF) designed for fixed-wing UAV navigation under GNSS-denied and GNSS-degraded scenarios. The proposed framework integrates an INS with magnetometer, barometer, pitot tube with pseudovanes, Visual-Inertial Odometry (VIO), and Computer Vision (CV) Convolutional Neural Network (CNN)-based position estimates within a unified estimation architecture. A comprehensive mathematical description of the ESEKF is provided, with special attention given to the modeling of sensor systematic errors, which are explicitly included in the augmented error-state vector. The estimator performance is evaluated through a flight experiment using a custom fixed-wing UAV platform, where the contribution of each aiding sensor is assessed via systematic sensor deactivation under GNSS outage conditions. Experimental results demonstrate the robustness and modularity of the proposed ESEKF architecture, highlighting its ability to maintain navigation accuracy despite the loss of individual aiding sensors (except barometer and CV). The presented framework consolidates previously fragmented sensor fusion approaches into a single, self-consistent navigation solution, providing both a practical implementation reference and a foundation for future extensions and tuning strategies.

Keywords: Navigation, Perception and Cognition, Autonomy
Abstract: Reliable backup localization for unmanned aerial vehicles (UAVs) operating in GNSS-denied nighttime conditions remains an open challenge due to the severe modality gap between daytime RGB maps and nighttime thermal imagery. This work presents a semantic reprojection framework for map-relative nighttime UAV localization by aligning segmented thermal observations with a globally referenced, semantically labeled 3D map constructed from daytime RGB data. Rather than relying on appearance-based correspondence, localization is formulated in a shared semantic domain and solved via a symmetric bidirectional reprojection objective with confusion-aware weighting to improve robustness under segmentation uncertainty. The approach is evaluated offline across 6.5,km of nighttime, real-world UAV flight trajectories in urban and semi-structured environments. Relative to RTK ground truth, the system achieves a bias-corrected mathrm{RMSE}_{2D} of 2.18,m and a median mathrm{RMSE}_{2D} of 1.52,m. Results show that localization performance is strongly correlated with the availability of semantic edge evidence and that large-error events are spatially localized to semantically ambiguous areas rather than uniformly distributed. These findings indicate that semantic reprojection offers a promising pathway toward globally referenced nighttime UAV localization using thermal imagery alone.

Keywords: Navigation, Path Planning, Swarms
Abstract: Safe swarm navigation in cluttered indoor environment requires long-horizon planning, reactive obstacle avoidance, and adaptive compliance. We propose ImpedanceDiffusion, a hierarchical framework that leverages image-conditioned diffusion-based global path planning with Artificial Potential Field (APF) tracking and semantic-aware variable impedance control for aerial drone swarms. The diffusion model generates geometric global trajectories directly from RGB images. These trajectories are tracked by an APF-based reactive layer, while a VLM-RAG module performs semantic obstacle classification with 90 % retrieval accuracy to adapt impedance parameters for mixed obstacle environments during execution. Two diffusion planners are evaluated: (i) a top-view long-horizon planner using single-pass inference and (ii) a first-person-view (FPV) short-horizon planner deployed via a two-stage inference pipeline on top view. Both planners achieve a 100 % trajectory generation rate across twenty static and dynamic experimental configurations and are validated via zero-shot sim-to-real deployment on Crazyflie 2.1 drones through the hierarchical APF-impedance control stack. The top-view planner produces smoother trajectories that yield conservative tracking speeds of 1.0-1.2 m/s near hard obstacles and 0.6-1.0 m/s near soft obstacles. In contrast, the FPV planner generates trajectories with greater local clearance and typically higher speeds, reaching 1.4-2.0 m/s near hard obstacles and up to 1.6 m/s near soft obstacles. Across 20 experimental configurations (100 total runs), the framework achieved a 92% success rate while maintaining stable impedance-based formation control with bounded oscillations and no in-flight collisions demonstrating reliable and adaptive swarm navigation.

Keywords: Airspace Management, Simulation, Interoperability
Abstract: The digital twin paradigm is increasingly recognized as a key enabler for advancing autonomy, safety, and operational reliability in unmanned aviation. This paper presents a digital twin‑based simulation framework designed for single and multi‑UAV operations, integrating offline and real‑time execution modes within a unified environment. The system combines high‑fidelity dynamic models of multiple UAV platforms with a MAVLink‑based communication layer that ensures seamless interoperability with ground control stations such as QGroundControl and Mission Planner. In addition, the framework incorporates an interface to a U‑Space Service Provider (USSP), enabling the evaluation of U‑Space services and procedures in realistic operational scenarios. This integration supports real‑time testing, hardware‑in‑the‑loop experiments, and the assessment of cooperative mission behaviors under varying traffic and airspace conditions. By providing a controllable and risk‑free environment, the digital twin reduces development time, cost, and safety concerns associated with real flight testing, while enabling both technical and operational performance evaluations. The paper describes the system architecture, key functionalities, and experimental capabilities, and outlines future research directions for enhancing digital‑twin‑enabled UAV and U‑Space operations.

Keywords: Airspace Management, Security, Manned/Unmanned Aviation
Abstract: Remote Identification (RID) is increasingly mandated to enhance drone traceability by broadcasting a UAV’s identity, position, and other metadata for ground-based monitoring. However, the transmitted location can be falsified by operators to conceal restricted-area violations or compromised by GPS spoofing, causing ground observers to receive deceptive positional data. To address this, we propose a ground-based cross-modal verification system that validates RID-reported locations using visual data from a calibrated RGB camera. Given the camera’s known position and the UAV’s physical dimensions (retrieved via drone ID or inferred in real time) the system infers the UAV’s range and compares it to the GPS-derived distance. Discrepancies exceeding an adaptive threshold are flagged as potential spoofing or manipulation events. For reproducibility, we release DroneR, a public dataset of UAV frames annotated with ground-truth range information. DroneR contains 208 images across two drone platforms, spanning distances of 3–50 m under varied lighting and background conditions, and includes scripted distance-error profiles that emulate spoofing scenarios. Experiments on DroneR demonstrate that the proposed verification system reliably detects both large and minor location falsifications, achieving a true positive rate of 0.8–0.91 with a corresponding false positive rate between 0.25 and 0.05. This performance supports consistent operational accuracy and practical deployability. The approach provides a complementary verification layer that strengthens cryptographic RID protections and enhances the security of UTM ecosystems. The code and dataset are publicly available at github.com/KU-USL/visual-location-authentication.

Keywords: Airspace Control, Control Architectures
Abstract: It has recently been shown that all physical parameters of an Incremental Nonlinear Dynamic Inversion (INDI) controller can be estimated onboard a multirotor within half a second, which is fast enough to do the full identification during a throw in the air. However, a robust method to tune outer loop gains for this feedback-linearizing INDI controller depending on the model parameters is still missing. This work presents the design of a robust gain-scheduled controller for attitude control of quadrotor, using an INDI-based inner loop with online identification of its system parameters. A gain-scheduled cascaded attitude controller with a feedforward filter is synthesized for a symmetric quadrotor using signal-based H∞ closed-loop shaping. The resulting controller exhibits good stability margins, with nonlinear simulations confirming effective tracking performance under uncertainty. Experimental evaluation is also conducted through flight tests with full online parameter identification. Even though the identified parameters during these tests are far outside the defined uncertainty range, acceptable flight performance comparable to simulation results is maintained for actuator time constants below 40 ms.

Keywords: Airspace Control, Aerial Robotic Manipulation
Abstract: The use of unmanned aerial vehicles (UAVs) for physical interaction and aerial manipulation introduces significant technical challenges. While quadrotors are widely adopted for these tasks, their reliance on purely vertical thrust fundamentally limits their energy efficiency during sustained transport missions. Hybrid tilt-rotor configurations decouple thrust vectoring from attitude control, at the cost of tightly coupled, nonlinear aerodynamics during transition. The fixed-wing tilt-rotor configuration investigated in this paper features independently tiltable left and right rotor pairs, decoupling pitch from forward acceleration, so the wing can operate at its optimal angle of attack. This paper presents a unified control architecture based on Incremental Nonlinear Dynamic Inversion (INDI) for this platform, specifically applied to tethered flight. Simulation results on a circular trajectory demonstrate both the controller's tracking accuracy and the efficiency of the platform: compared to a baseline quadrotor, the tilt-rotor reduces specific power consumption by 47%, and the tethered configuration by 38% while maintaining precise tracking through transition under the tether disturbances.

Keywords: See-and-avoid Systems, UAS Applications, Airspace Control
Abstract: High-Andean wetlands are critical for water supply in Ecuador, but their monitoring requires beyond-visual-line-of-sight (BVLOS) UAV operations in harsh, obstacle-rich environments. Fixed-wing and VTOL platforms are suitable for large-area monitoring but lack affordable obstacle avoidance systems adapted to high-speed flight. This work proposes a deep-learning-based obstacle detection framework for high-speed VTOL UAVs operating at up to 20 m/s. The proposed network, based on an enhanced SSD architecture with binary mask segmentation, estimates obstacle dimensions (height and width) and distance at ranges up to 200 m, enabling early detection compatible with aerodynamic maneuvering constraints. For safe development and validation, a high-fidelity digital twin was implemented in ROS and Gazebo, integrating a five-motor VTOL model with ArduPilot and QGroundControl. The digital twin enables controlled simulation, dataset generation, and algorithm evaluation without operational risk. Results show reliable long-range obstacle detection across multiple distance intervals, supporting future integration with collision avoidance systems for high-speed environmental-monitoring UAVs.