Autonomous UAVs operating in shared uncontrolled airspace must safely navigate to mission destinations while maintaining separation from surrounding aerial traffic. This task becomes particularly challenging under adversarial disturbances such as persistent Global Navigation Satellite System (GNSS) spoofing, which systematically adds a bias to the navigation estimates and induces a distribution shift between training and deployment observations, thus degrading previously learned navigation policies. Existing reinforcement learning (RL) methods generally assume similar training and deployment conditions, while robust RL approaches account only for predefined distribution shifts or uncertainty sets during training. Hence, previously unseen spoofing characteristics can invalidate learned value estimates, often requiring computationally expensive policy retraining. To address this limitation, we propose an offline decision-theoretic framework that augments existing robust and nominal navigation policies with a stochastic meta-policy. The framework quantifies policy resilience using adversarially induced critic-value distribution shifts, which serve as an effective proxy for mission-level performance and are transformed into offline Bayesian posterior beliefs using discounted Thompson sampling (DTS). During deployment, these fixed posterior beliefs enable lightweight uncertainty-aware switching among policies without requiring retraining. Extensive simulations involving heterogeneous aerial traffic under GNSS spoofing demonstrate up to a 70% improvement in mission success, more than a 50% reduction in conflicts, and reliable navigation under severe observation corruption. Broadly, the proposed framework introduces a general decision layer for resilient deployment of robust policies under previously unseen adversarial disturbances.
Aeronautical decision-making by pilots and air traffic controllers (ATCOs) is often complex and requires integrating multiple pieces of information from the environment. It necessitates emotional regulation to make the appropriate aeronautical decision based on the information perceived from the scenario. Hence, recognising the emotions of pilots and ATCOs is a prerequisite to designing appropriate support for emotion-aware flight operations. Nevertheless, generic speech emotion recognition models might face challenges in recognising emotions from aviation communications with a specific phraseology. In this regard, we first introduce the Cranfield Aviation Speech Emotion Set (CASES), a corpus tailored for speech emotion recognition in aviation. This corpus comprises 30 diverse utterances recorded from 20 pilots and ATCOs with multiple accents across seven basic emotions. We then developed a deep learning model based on long short-term memory (LSTM) and CASES to recognise emotions in aviation communications. A data augmentation technique has been applied to enhance the model's prediction performance. Our proposed model achieves a test accuracy of 84.52 %, outperforming models trained with generic datasets and demonstrating the need for a tailor-made corpus in aviation applications. Shapley additive explanations (SHAP) are further leveraged to reveal feature importance in the emotion recognition process, thereby enhancing the model’s explainability. Overall, the model supports trustworthy emotion-aware flight operations for minimising emotional impacts on aeronautical decision-making.
Reconfigurable manufacturing requires robotic workcells that can adapt to changing products, processes, and equipment without extensive redesign or manual reprogramming. This paper presents the R3M framework, a ROS 2 driven integration architecture that connects model based manufacturing knowledge with execution level robotic control. Product, process, and equipment information is formalised through UML domain models and serialised in AutomationML(AML), enabling automated correspondence between assembly requirements, available skills, and executable recipes. The framework integrates automated programme generation, reinforcement learning based recipe optimisation, and a modular CAD informed six degree of freedom perception layer to support both technical and semantic interoperability across simulated and physical workcells. The approach is evaluated through Cube Kitting and Cylinder Stacking use cases implemented on distinct robotic platforms, including ABB and Universal Robots systems. Experimental results show high reliability, with environment launch performance reaching up to 99.93%, standard skill sequences achieving 100% execution success, and full use case trials exceeding 97% success in simulation and reaching 100% on physical hardware. These findings demonstrate that R3M provides a scalable foundation for adaptive robotic manufacturing, reducing programming effort while supporting modular substitution, robust perception, and simulation to real deployment.
Floating photovoltaic (FPV) systems offer a promising route for expanding solar generation in coastal and offshore water bodies. As arrays grow to larger multi-floater configurations, the wave-induced motions, inter-module connector forces, and mooring loads become complex. This study presents high-fidelity CFD simulations of 3×3 and 4×4 rope-mesh FPV arrays subjected to regular wave loading. Each module was modelled as an independent six-degree-of-freedom rigid body with elastic spring connectors and a mooring system, enabling insights into floater motions, connector forces, and mooring loads at the individual panel level across 16 wave conditions. The RAOheave exhibited front-to-rear attenuation driven by wave sheltering, with the front row heave increasing and the rear row heave decreasing at shorter relative wavelengths, progressively increasing the range of motion contrast observed across the array. Pitch amplification was observed in the 4×4 array at λ/L ≈ 1.43 and H/h = 2.14. In the 3×3 array, the highest streamwise connector forces occur between the middle and rear floaters, whereas in the 4×4 array, they shift to the connectors between the interior unmoored floater rows, indicating that the critical load-bearing connector location moves from the array boundary toward the interior as the array grows. Fast Fourier Transform (FFT) decomposition of the mooring force time histories showed a significant second-harmonic contribution, with F₂/F₁ ratios of 0.29–0.81 across all tested conditions, indicating that mooring systems designed solely on peak tension could progressively underestimate the secondary loading and fatigue as FPV arrays scale to larger configurations.
Wire-based directed energy deposition (DED) additive manufacturing (AM) uses an intense energy source, such as an electric arc, laser, and electron beam, to melt metal wire feedstock and deposit a structural part layer by layer. This emerging manufacturing process is advantageous thanks to its large-scale deposition capacity, high efficiency of material and energy usage, and wide applicability to different industrial applications. However, research is still needed to enhance the process productivity and part quality. Wire preheating is a feasible method to significantly enhance deposition rate. It can also help reduce heat input of the main energy source, inhibit pore formation and refine grains, thereby enhancing mechanical properties of the deposited part. Induction heating (IH) is a highly controllable non-contact heating method suited for rapidly and precisely preheating the wire feedstock to a target temperature. In addition, compared with conventional weld wire preheating methods such as resistance heating and bypass heating, IH avoids magnetic blow and is applicable to most metals with flexible setup. However, IH-based preheating of moving wire feedstock is complicated and underexplored for AM applications. In this study, to understand the complex electromagnetic heating mechanism, a multiphysics finite element model of coupled electromagnetic and thermal fields is developed based on the formulation in Eulerian frame, which improves the computational efficiency by 80.9 % compared to the model in Lagrangian frame. Furthermore, in the case of feedstock passing through a stationary magnetic field at a constant wire feed speed, a more efficient steady-state approach is proposed with 98.9 % computational time saving than the transient model. The temperature predictions by the models were validated by thermocouple measurement in an experiment. A range of coil geometries and setups were evaluated using the developed efficient model, revealing the coil effects on the wire preheating temperature and energy consumption.
