Animal ecology data collection using drones represents a substantial investment of time, expertise, and financial resources. Yet most existing datasets serve only a single research community, limiting interdisciplinary reuse. We propose a unified drone dataset standard, FAIR^2 Drones, that bridges ecology, robotics, and computer vision by building on existing FAIR and AI-ready data frameworks while adding essential platform metadata and annotation specifications. Our standard enables datasets to simultaneously support ecological analysis, robotics algorithm development, and computer vision benchmarking. We provide open-source validation tools, reference implementations, and multimodal extensions linking drone imagery with complementary sensors such as camera traps, GPS, and acoustics. By standardizing metadata across disciplines, this framework maximizes the scientific return on investment for costly field deployments and accelerates cross-domain collaboration in environmental monitoring.

Real-time wildlife detection in drone imagery supports critical ecological and conservation monitoring. However, standard detection models like YOLO often fail to generalize across locations and struggle with rare species, limiting their use in automated drone deployments. We present MMLA, a novel multi-environment, multi-species, low-altitude drone dataset collected across three sites (Ol Pejeta Conservancy and Mpala Research Centre in Kenya, and The Wilds in Ohio), featuring six species (zebras, giraffes, onagers, and African wild dogs). The dataset contains 811K annotations from 37 high-resolution videos. Baseline YOLO models show performance disparities across locations while fine-tuning YOLOv11m on MMLA improves mAP50 to 82

Real-time wildlife detection in drone imagery is critical for numerous applications, including animal ecology, conservation, and biodiversity monitoring. Low-altitude drone missions are effective for collecting fine-grained animal movement and behavior data, particularly if missions are automated for increased speed and consistency. However, little work exists on evaluating computer vision models on low-altitude aerial imagery and generalizability across different species and settings. To fill this gap, we present a novel multi-environment, multi-species, low-altitude aerial footage (MMLA) dataset. MMLA consists of drone footage collected across three diverse environments: Ol Pejeta Conservancy and Mpala Research Centre in Kenya, and The Wilds Conservation Center in Ohio, which includes five species: Plains zebras, Grevy's zebras, giraffes, onagers, and African Painted Dogs. We comprehensively evaluate three YOLO models (YOLOv5m, YOLOv8m, and YOLOv11m) for detecting animals. Results demonstrate significant performance disparities across locations and species-specific detection variations. Our work highlights the importance of evaluating detection algorithms across different environments for robust wildlife monitoring applications using drones.

This paper presents a range control approach for implementing hard geofencing for unmanned air vehicles (UAVs), and especially remotely piloted versions (RPVs), via a proposed anticipatory range calculator. The approach employs turning circle intersection tests that anticipate the fence perimeter on approach. This ensures the vehicle turns before penetrating the geofence and remains inside the allowable operational airspace by accounting for the vehicles’ turning dynamics. Allowance is made for general geozone shapes and locations, including those located at the problematic poles and meridians where nonlinear angle mapping is dealt with, concave geozones, narrow corners with acute internal angles, and transient turn dynamics. The algorithm is shown to prevent any excursions using a high-fidelity simulation of a small remotely piloted vehicle. The algorithm relies on a single tuning parameter which can be determined from the closed-loop rise time in the aircraft’s roll command tracking.

This paper describes the development and performance verification of a unique wind tunnel model and traversing rig for testing the use of active flow control technology as a novel control effector. The rig allows the model to be set within the University of Bristol’s 7′ × 5′ low speed tunnel and moved through a wide �� and �� range (−10◦ to 35◦ and ±25◦ respectively) whilst supplying blowing air to the model. The development and verification of taring procedures and methods for converting ‘in-tunnel’ measured data to ‘free air’ conditions are described. A set of tests featuring a generic combat aircraft with a 65◦ leading edge is used in order to demonstrate the capabilities of the system. This work serves as a precursor to future publications where further test results and comparisons with CFD simulations will be presented.

This paper discusses a method of determining the minimum safe altitude of an uncrewed aerial vehicle (UAV) at any point within a designated airspace by conducting a glide reachability analysis. Recently, fixed-wing UAVs are more regularly deployed near population centers and in extreme environments, requiring increasingly robust emergency systems and planning. The long-ranges and adverse terrain associated with monitoring the Volcan de Fuego in Guatemala by a team from the University of Bristol (UoB) increases the likelihood that motor failure would result in the aircraft being unable to Return To Home (RTH) and impossible to retrieve. A method for delineating a boundary representing the minimum safe altitude required for the aircraft to safely glide to the airfield in the event of a motor failure was developed within MATLAB, defined by the UAV's minimum glide angle in wind. This model was subsequently compared with flight data from UoB missions around Fuego to better improve its accuracy and analyze the limitations of the missions.

Bagana is a remote, highly active volcano, located on Bougainville Island in southeastern Papua New Guinea. The volcano has exhibited sustained and prodigious sulfur dioxide gas emissions in recent decades, accompanied by frequent episodes of lava extrusion. The remote location of Bagana and its persistent activity have made it a valuable case study for satellite observations of active volcanism. This remoteness has also left many features of Bagana relatively unexplored. Here, we present the first measurements of volcanic gas composition, achieved by unoccupied aerial system (UAS) flights through the volcano's summit plume, and a payload comprising a miniaturized MultiGAS. We combine our measurements of the molar CO 2 /SO 2 ratio in the plume with coincident remote sensing measurements (ground‐ and satellite‐based) of SO 2 emission rate to compute the first estimate of CO 2 flux at Bagana. We report low SO 2 and CO 2 fluxes at Bagana from our fieldwork in September 2019, ∼320 ± 76 td −1 and ∼320 ± 84 td −1 , respectively, which we attribute to the volcano's low level of activity at the time of our visit. We use satellite observations to demonstrate that Bagana's activity and emissions behavior are highly variable and advance the argument that such variability is likely an inherent feature of many volcanoes worldwide and yet is inadequately captured by our extant volcanic gas inventories, which are often biased to sporadic measurements. We argue that there is great value in the use of UAS combined with MultiGAS‐type instruments for remote monitoring of gas emissions from other inaccessible volcanoes.

Wildfires affect countries worldwide as global warming increases the probability of their appearance. Monitoring vast areas of forests can be challenging due to the lack of resources and information. Additionally, early detection of wildfires can be beneficial for their mitigation. To this end, we explore in simulation the use of swarms of uncrewed aerial vehicles (UAVs) with long autonomy that can cover large areas the size of California to detect early stage wildfires. Four decentralised control algorithms are tested: (1) random walking, (2) dispersion, (3) pheromone avoidance and (4) dynamic space partition. The first three adaptations are known from literature, whereas the last one is newly developed. The algorithms are tested with swarms of different sizes to test the spatial coverage of the system in 24 h of simulation time. Best results are achieved using a version of the dynamic space partition algorithm (DSP) which can detect 82% of the fires using only 20 UAVs. When the swarm consists of 40 or more aircraft 100% coverage can also be achieved. Further tests of DSP show robustness when agents fail and when new fires are generated in the area.

Capturing vertical profiles of the atmosphere and measuring wind conditions can be of significant value for weather forecasting and pollution monitoring however, collecting such data can be limited by current approaches using balloon-based radiosondes and expensive ground-based sensors. Multirotor vehicles can be significantly affected by the local wind conditions, and due to their under-actuated nature, the response to the flow is visible in the changes in the orientation. From these changes in orientation, wind speed and direction estimates can be determined, allowing accurate estimation with no additional sensors. In this work, we expand on and improve this method of wind speed and direction estimation and incorporate corrections for climbing flight to improve estimation during vertical profiling. These corrections were validated against sonic anemometer data before being used to gather vertical profiles of the wind conditions around Volcan De Fuego in Guatemala up to altitudes of 3000 m Above Ground Level (AGL). From the results of this work, we show we can improve the accuracy of multirotor wind estimation in vertical profiling through our improved model and some of the practical limitations of radiosondes that can be overcome through the use of UAS in this application.

In recent years, the air-drop launch technology of near-space UAVs has attracted much attention. Between downfall from the carrier and the flight control system's initiation, the UAV presents free-fall movement. This free-fall process is very important for the control effect of the flight control system and is also crucial for the safety of the UAV and the carrier. Focus is required on two important dynamic parameters of the UAV: the moment of inertia and the center of mass position. In this paper, we used a quasi-steady model proposed by predecessors to address the flat-plate falling problem with modifications to describe the freely falling motion of the wing. Computational fluid dynamics (CFD) were used to simulate the free-fall movement of the wing with various parameters, and the wing release behavior was analyzed to check the quasi-steady model. Research shows that the movement characteristics of the falling wing are mostly reflected in the longitudinal plane, and the developed quasi-steady analytical model can more accurately describe the dynamic behavior of free-fall to some extent. By using CFD methods, we further investigated the aerodynamic performance of the free-fall wing. The results show that the wing mainly presents tumbling and fluttering motion. Changing the moment of inertia around the tumbling axis changes the tumbling frequency and the time point as the wing enters tumbling. In contrast, changing the position of the center of mass significantly changes the form of falling and makes the free-fall motion more complex. Therefore, it is necessary to carefully configure the center of mass in the UAV design process.

Reinforcement learning has been used on a variety of control tasks for drones, including, in previous work at the University of Bristol, on perching manoeuvres with sweep-wing aircraft. In this paper, a new aircraft model is presented representing flight up to very high angles of attack where the aerodynamic models are highly nonlinear. The model is employed to develop high-alpha manoeuvres, using reinforcement learning to exploit the nonlinearities at the edge of the flight envelope, enabling fixed-wing operations in tightly confined spaces. Training networks for multiple manoeuvres is also demonstrated. The approach is shown to generate controllers that take full advantage of the aircraft capability. It is suggested that a combination of these neural network-based controllers, together with classical model predictive control, could be used to operate efficiently within the low alpha flight regime and, yet, respond rapidly in confined spaces where high alpha, agile manoeuvres are required.

Multi-rotor and fixed-wing Uninhabited Air Vehicles (UAVs) have been used extensively in recent years for atmospheric studies. Multirotors can hover and climb/descend vertically hence, they can map the vertical profiles of the atmosphere faster than fixed-wings which lead to a higher temporal resolution in the data. As opposed to this, fixed-wing configurations are more energy efficient, especially during the descend stage where they can glide down and consume negligible amounts of energy. In this paper, a novel approach for the minimal energy and controlled descent of quadrotors by free-falling has been introduced. Two sets of experiments have been conducted and presented: 1. wind-tunnel testing of a single motor, electrical speed controller, and propeller combination, and 2. uncontrolled free-fall of a small quadrotor (built of off-the-shelf components) in Guatemala. In these experiments, the dynamic stability, controllability, energy consumption, and regeneration aspects have been investigated. In the uncontrolled free-fall trial, it has been shown that the current consumption drops to nearly zero amps and a terminal velocity of 18.5 m/s was reached. It has been found that uncontrolled free-falling descent, from 330 to 120 m above ground level, has resulted in 98% improvement in energy consumption compared to a powered descent at 5 m/s. Furthermore, based on the wind tunnel results, it has been proven that it is possible to recharge the flight batteries during a fall.

High-altitude pseudo-satellites are an expanding focal area of the aerospace industry which require new technologies and manufacturing processes to reduce weight and increase efficiency with the aim of increasing endurance. One such process has resulted in the occurrence of small deformities along the leading edge of a lightweight unmanned aerial vehicle structure with the application of its skin, which may have a detrimental impact on its performance and efficiency. This paper focuses on the effects of these manufacturing deformities on the aerodynamic performance of the vehicle’s aerofoil when operating in low Reynolds number flow with the intention of identifying any detrimental flow variation. This analysis is achieved by comparing the lift curve, drag polar and pressure coefficient of both the deformed and undeformed cases of two aerofoils: a SG6042 and a GOE 523. This is accompanied with an examination of the local flow conditions scrutinising the near-wall y+ and turbulent kinetic energy calculations. The investigation finds that in two-dimensional flow, the deformities replicate the effects of transition trips in the shrinking or elimination of laminar separation bubbles. At Reynolds numbers below 250,000, the deformities reduce the net drag while leaving the lift largely unaffected. However, as a result, there is a slight shift in the minimum power condition in the order of 8% which would produce some performance loss for power efficiency and endurance.

Volcanic ash is an aerial hazard to both aircraft and local populations, hence efforts to monitor and quantify its presence are underway in many regions, both to protect airspace users and as a modeling tool for volcanology. One direct approach is to acquire in-situ ash samples using an unmanned vehicle, but this introduces possible bias in the samples due to the collector system, especially across the full range of ash particle sizes. This work explores an open impact type design of miniaturized airborne ash collector and quantifies the influence of the collector geometry on the measured particle size distribution. Results indicate that the distribution of sizes measured can be strongly influenced by the collector design, owing to the influence of the geometry on the aerodynamics and hence the particle trajectories. Comparisons to experimental flight data illustrate that the simulations provide a representative model, and that a design featuring a bluff fore-body separates and mixes the flow, making it more capable of capturing small particles than a planar surface in isolation.

This work presents an open-source Python module for flight dynamics modelling, with the core numerical operations implemented in Rust for improved performance. The primary use case is as part of machine learning research, where Python is the dominant language. Previous machine learning work by the authors had begun to show the performance limitations of a pure Python flight dynamics model. The module presented here began development in order to accelerate future models. To evaluate the module, a simple model was implemented using it and an equivalent was created for JSBSim. This was then used to benchmark performance and compare interfaces and workflows. For the simple model used, performance of the new module is shown to be comparable to JSBSim, and indeed faster when extracting state at each timestep from a Python wrapper. In future machine learning work, this improved performance will allow more training time for machine learning models, resulting in better trained agents capable of performing more complex manoeuvres.

Reinforcement learning has previously been applied to the problem of controlling a perched landing manoeuvre for a custom sweep-wing aircraft. Previous work showed that the use of domain randomisation to train with atmospheric disturbances improved the real-world performance of the controllers, leading to increased reward. This paper builds on the previous project, investigating enhancements and modifications to the learning process to further improve performance, and reduce final state error. These changes include modifying the observation by adding information about the airspeed to the standard aircraft state vector, employing further domain randomisation of the simulator, optimising the underlying RL algorithm and network structure, and changing to a continuous action space. Simulated investigations identified hyperparameter optimisation as achieving the most significant increase in reward performance. Several test cases were explored to identify the best combination of enhancements. Flight testing was performed, comparing a baseline model against some of the best performing test cases from simulation. Generally, test cases that performed better than the baseline in simulation also performed better in the real world. However, flight tests also identified limitations with the current numerical model. For some models, the chosen policy performs well in simulation yet stalls prematurely in reality, a problem known as the reality gap.

Vision processing for control of agile autonomous robots requires low-latency computation, within a limited power and space budget. This is challenging for conventional computing hardware. Parallel processor arrays (PPAs) are a new class of vision sensor devices that exploit advances in semiconductor technology, embedding a processor within each pixel of the image sensor array. Sensed pixel data are processed on the focal plane, and only a small amount of relevant information is transmitted out of the vision sensor. This tight integration of sensing, processing, and memory within a massively parallel computing architecture leads to an interesting trade-off between high performance, low latency, low power, low cost, and versatility in a machine vision system. Here, we review the history of image sensing and processing hardware from the perspective of in-pixel computing and outline the key features of a state-of-the-art smart camera system based on a PPA device, through the description of the SCAMP-5 system. We describe several robotic applications for agile ground and aerial vehicles, demonstrating PPA sensing functionalities including high-speed odometry, target tracking, obstacle detection, and avoidance. In the conclusions, we provide some insight and perspective on the future development of PPA devices, including their application and benefits within agile, robust, adaptable, and lightweight robotics.

All radiological measurements acquired from airborne detectors suffer from the issues of geometrical signal dilution, signal attenuation and a complex interaction of the effective sampling area of the detector system with the 3D structure of the surrounding environment. Understanding and accounting for these variables is essential in recovering accurate dose rate maps that can help protect responding workforces in radiologically contaminated environments. Two types of terrain-cognisant methods of improving source localisation and the contrast of airborne radiation maps are presented in this work, comprising of ‘Point Cloud Constrained 3D Backprojection’ and ‘Point Cloud Constrained Randomised Kaczmarz Inversion’. Each algorithm uses a combination of airborne gamma-spectrometry and 3D scene information collected by UAS platforms and have been applied to data collected with lightweight, simple (non-imaging) detector payloads at numerous locations across the Chornobyl Exclusion Zone (CEZ). Common to both the algorithms is the projection of the photopeak intensity onto a point cloud representation of the environment, taking into account the position and orientation of the UAS in addition to the 3D response of the spectrometer. The 3D Backprojection method can be considered a relatively fast method of mapping of through proximity, in which the measured photopeak intensity is split over the point cloud according to the above factors. It is an additive technique, with each measurement increasing the overall magnitude of the radiation field assigned to the survey area, meaning that more measurements continues to increase the total radiation of the site. The total measured intensity of the solution is then normalised according to the time spent in proximity to each point in the scene, determined by splitting and projecting the nominal measurement time at each survey point over the point cloud according to the distance from the survey position. Thus accounting for sampling biases during the survey. The inversion approach adapts algorithms routinely used in medical imaging for the unconstrained world in which the detector is no longer completely surrounding the subject/target. A forward projection model, based on the contribution of distant point sources to the detector intensity, is used to determine the relationship between the full set of measurements and the 3D scene. This results in a hypercube of linear equations where it is assumed every point in the scene contributes to the measured intensity. The algorithm randomly adds measurements from within the aerial set and back-projects this onto the point cloud, with the initial state of the solution set to emit no radiation. After a given number of iterations, the fit of the current solution to the original measurements is assessed though a least squares method and updated when this produces a fit better than the current best estimate. This continues to happen until a minimum value is reached before the divergence of the system, representing the most confident solution. Based on examples from both simulations and real world data, the improvement in contrast of airborne maps using this inversion method can make them equivalent to ground-based surveys, even when operating at 20 m AGL and above.