Open research questions in UAV Applications and Optimization

What the literature leaves open

• FUTURE WORK Future research will focus on methods to estimate the UAV or gimbal camera roll/pitch angles, thereby extending the positioning capability to 6-DoF, which would provide a more complete representation of the UAV’s orientation in space.

  SWA-PF: Semantic-weighted adaptive particle filter for memory-efficient 4-DoF UAV localization in GNSS-denied environments · 2026 · DOI

• The paper introduced QuantumFed-ARIS, which is an inno- vative framework that combines quantum machine learning, federated continual learning, digital twin technology, and intent-based networking to manage autonomously recon- figurable intelligent surface (RIS)-equipped UAV constel- lation networks in heterogeneous 6G communications. The framework proposed four tightly coupled innovations: (i) a Quantum Variational Policy Network which makes use of quantum superposition to explore the joint RIS phase-shift and UAV trajectory solution space exponentially efficiently; (ii) a Federated Elastic Weight Consolidation protocol that allows privacy-preserving collaborative learning without catastrophic forgetting under non-stationary deployment conditions; (iii) a Real-Time Digital Twin Engine which. QuantumFed-ARIS was demonstrated to deliver 52% spectral efficiency gain, 41% energy savings, 99.7% com- munication reliability and 78 × faster convergence than five state-of-the-art baselines in extensive experiments across twelve different deployment scenarios, including dense urban, rural broadband, maritime IoT, LEO satellite-terres- trial integration, disaster relief, and industrial automation environments, and reduces cat The quantum variational policy uses 120 parameters to be trained, three orders of magnitude less than classical counterparts, and can be implemented with resource-constrained UAV platforms. Future research opportunities involve: (1) implement- ing QuantumFed-ARIS on real quantum hardware running hardware-aware noisy quantum interactions and involv- ing hardware-aware noise mitigation, no longer based on simulation; (2) implementing quantum entanglement-based communication protocols between UAVs to achieve better coordination; (3) implementing large language models into the HIAS component to achieve natural-language intent parsing with zero-shot generalization to new types. Author contributions Sameer conceived and designed the study, de- veloped the deep reinforcement-learning (DRL) framework of UAV- RIS system, conducted software simulations, analyzed the results, and wrote the initial draft of the manuscript. Girish J supervised the re- search, provided guidance throughout the project, critically revised the manuscript, assisted in data analysis, result interpretation, and manu- script formatting. Both authors reviewed the final manuscript, and pro- vided intellectual input during manuscript revision. All authors have read and approved the final version of the manuscript and agree to be accountable for all aspects of the work. Data availability The data supporting the findings of this study are available from the corresponding author upon reasonable request.

  Quantum fed-ARIS: quantum-enhanced federated continual learning for autonomous RIS-UAV constellation networks with digital twin-guided predictive optimization · 2026 · DOI

• This paper investigates the problem of remote IoT task offloading and resource allocation for LEO satellites under a dual-layer heterogeneous network collaborative architecture, considering their high mobility and dynamically changing channel environment, with the goal of minimizing the total computation cost for UDs. To address the complex MINLP problem, the optimization problem is decomposed into different subproblems, which are solved individually. An AOTORA algorithm based on the space-air dual-layer network architecture is proposed. Simulation results show that the proposed scheme outperforms other benchmark schemes in both small-scale and large-scale scenarios. In terms of the average total computation cost for users, it reduces costs by 12.48%, 25.8%, 11.63%, and 6.84% compared with OLS, RO, EBA, and DRO, respectively. Despite these achievements, this work is based on a static resource allocation framework within a single time slot and does not account for the dynamic randomness of tasks and the ultra-large-scale networks characteristic of IoT scenarios. Therefore, in our future work, we will study a multi-time-slot dynamic optimization framework that incorporates task data dependencies and user mobility models to enhance long-term computation adaptability.

  Joint task offloading and resource allocation for remote IoT in space-air-ground networks architecture · 2026 · DOI

• Finally, it identifies open challenges and future directions in realistic channel modeling, energy-neutral operation, benchmarking, reproducibility, scalable and trustworthy AI, security, privacy, hardware validation, and integration with RIS, MEC, digital twins, and 6G technologies.

  AI-Empowered UAV-Assisted Backscatter Localization and ISAC for Zero-Energy IoT: A Comprehensive Survey · 2026

• A noticeable performance gap is also observed between the fixed partition with trajectory optimization and the static RIS scheme, demonstrating that static partitioning or static control alone is insufficient to fully exploit system potential.

  Joint design of UAV energy harvesting and heterogeneous user communications based on dynamic STAR-RIS partitioning · 2026 · DOI

• The results demonstrate that path loss alone is insufficient to describe UAV communication channels, as CIR and PDP provide additional insight into multipath propagation and delay-domain behaviour.

  Communication Channel Modelling of Unmanned Aerial Vehicles · 2026

• Finally, we summarize testbeds, datasets, and evaluation metrics, and provide representative use cases to illustrate how the proposed framework translates application requirements into practical task-driven optimization designs, together with open challenges and a roadmap toward scalable and trustworthy LAE deployment.

  Collaborative Air-Ground Sensing, Communication, Computing, Storage, and Intelligence for Low-Altitude Economy · 2026

• In this study, a delivery network with drones was designed at three levels: warehouses, charging stations, and customers, focusing on location, allocation, and charging decisions. This network was modeled as a robust mixed-integer programming problem to minimize total network costs. The proposed model, while maintaining linearity, is simultaneously robust against fluctuations in demand and drone energy consumption. A robust approach with budget uncertainty was used for demand fluctuations, and a fuzzy-robust approach with the Me criterion and fuzzy absolute deviation was used for drone energy consumption fluctuations. Thus, the proposed model maintains its efficiency and robustness in the worst-case operational scenarios. Several numerical examples and sensitivity analyses were performed to demonstrate the model’s performance and to analyze its behavior under fluctuations in various parameters. Sensitivity analyses were performed with respect to the budget parameters of demand uncertainty, backup capacity, unmet demand penalty, and environmental coefficient. The budget parameter of uncertainty directly affects the model’s conservatism and, consequently, the network costs. Increasing ΓD increases the network cost by 31.16% on average while allowing the design of a network that is resilient to worst-case scenarios. Increasing ΓD changes the network structure, indicating the model’s conservatism and an attempt to optimize resource utilization. Available backup capacity plays an important role in reducing the effects of demand fluctuations. Activating the backup capacity comes with a certain cost, but the absence of this capacity leads to an even further increase in costs. If the unmet demand penalty increases, there is a greater incentive to activate backup capacity and reduce the risk of non-supply. Furthermore, when demand increases, it is more optimal to add backup capacity to warehouses than to establish a new one, and this approach is less costly for the network. An increase in uncertainty about drone energy consumption directly increases the network’s total cost. The amount of this increase depends strongly on the decision maker’s risk tolerance level (λ). Increasing λ (more optimism) means accepting higher risk and, consequently, higher costs. The results show that network costs increase by an average of 13.59%. To address this increase in 1 3Process Integration and Optimization for Sustainability costs, the model considers adding more charging stations and adjusting the optimal mix of warehouses and charging stations. To demonstrate the practical application of the proposed model in the real world, a case study of the Digikala Company was presented, and the optimal locations for warehouses and charging stations were determined.

  Robust Optimization of Drone Delivery Networks: Integrated Location-Allocation and Charging Decisions with Fuzzy-Robust Energy Consumption under Demand Uncertainty · 2026 · DOI

• The trajectory-planning algorithm is built on the principle of adaptive maneuvering: the UAV plans a route through a restricted area while maintaining a critical distance $D$ from objects whose exact coordi- nates remain unknown to the operator.

  UAV routing model with information disclosure constraint · 2026 · DOI

• Future research should focus on developing predictive clustering models that employ machine learning to evaluate user movement patterns and service demands to enable proactive UAV deployment and resource allocation.

  Blockchain-Enabled Clustering for Dynamic Resource Allocation and Task Offloading in UAV-Assisted MEC for 5G Network Slicing · 2026 · DOI

• FUTURE RESEARCH DIRECTIONS Despite the robustness of MeshRelay, several open challenges remain for decentralized mobile SOS: 1) Security and Sybil Attacks: Developing lightweight cryptographic signatures to prevent malicious nodes from injecting "Fake SOS" messages into the mesh.

  A Comprehensive Survey on Decentralized Emergency Communication Frameworks: Bridging the Gap Towards Infrastructure-Independent Mobile SOS · 2026 · DOI

• In this paper, we presented AirFogSim, a simulation platform that contributes to addressing the challenges of compu- tation offloading in UAV-integrated VFC. Compared with current simulators, the proposed AirFogSim offers a more comprehensive and realistic simulation environment, focusing on the unique characteristics of UAVs and VFC in multiple layers, and providing several key missions in this field. We also demonstrated the capabilities of AirFogSim through a case study of computation offloading in VFC. The results show that AirFogSim can effectively simulate the complex interactions between UAVs and vehicles. 15 A PREPRINT - SEPTEMBER 5, 2024 Future work includes enriching AirFogSim with more diverse missions and robust security models and applying the platform to a broader range of applications in ITS. Our aim is to continuously refine AirFogSim, making it an increasingly effective tool for the research community and contributing to the evolution of intelligent transportation systems.

  AirFogSim: A Light-Weight and Modular Simulator for UAV-Integrated Vehicular Fog Computing · 2026 · DOI

• This research presented a memory-efficient and residual-inspired deep learning frame- work for high-fidelity intrusion detection in UAV swarm networks. Addressing the growing need for robust cybersecurity in aerial communication systems, the proposed Residual 1D-CNN effectively captured complex temporal-spatial dependencies inher- ent in UAV network traffic. The architecture integrated three key design elements, residual feature reinforcement, adaptive class weighting, and hierarchical convolutional filtering, to enhance detection precision, model stability, and computational efficiency. ARTICLE IN PRESS ARTICLE IN PRESS ACCEPTED MANUSCRIPT The proposed model was trained and validated on the UAVIDS-2025 dataset, which reflects realistic UAV communication patterns simulated using the NS-3.24 environ- ment under IEEE 802.11ac and AODV protocols. Through systematic preprocessing, normalization, and class imbalance handling, the framework achieved exceptional results, with an overall accuracy of 99.71%, a macro F1-score of 0.9971, and an ROC- AUC of 0.9999. These results were further supported by robustness tests, including ablation and statistical significance analyses, demonstrating the superior performance of the proposed approach compared to CNN, RNN, LSTM, and ANN baselines. Furthermore, memory and computational profiling confirmed that the model maintains high detection accuracy while remaining lightweight and deployable in resource-constrained UAV environments. This balance between performance and efficiency underscores the framework’s potential for real-time threat detection in aerial networks, contributing meaningfully to the advancement of autonomous UAV cybersecurity and intelligent defense mechanisms. Although the proposed framework achieves state-of-the-art performance, several promising avenues remain open for further exploration. Future work could extend this research by integrating federated or decentralized learning paradigms to enhance privacy-preserving UAV collaboration without requiring centralized data aggregation. Moreover, dynamic adversarial training and reinforcement learning-based adapta- tion could be incorporated to improve resilience against evolving and unseen attack patterns in real-world UAV missions. Additionally, evaluating the framework under diverse flight conditions, communica- tion standards (e.g., IEEE 802.11ax or 5G-enabled UAV networks), and cross-domain datasets would strengthen its generalizability. Finally, lightweight model compression and quantization strategies could further optimize energy consumption for edge or onboard UAV deployment. These future directions aim to evolve the proposed sys- tem into a fully adaptive, secure, and interpretable intrusion detection solution for next-generation aerial intelligence ecosystems.

  Residual-aware lightweight deep learning framework for high-fidelity intrusion detection in UAV swarm networks · 2026 · DOI

• The training was performed entirely in simulation with a 2D discrete environment. Transfer learning from simulation to real physical drone platforms, robustness to sensor noise, actuator delays, and performance degradation due to sim-to-real gap in autonomous drone navigation using DQN has not been addressed.

  Optimized Autonomous Drone Navigation Using Deep Q-Network Based Reinforcement Learning · 2026 · DOI

• The paper does not specify grid environment sizes tested or how DQN performance scales with increasing grid dimensions (from the NxN notation used). Scalability to large navigation spaces (e.g., 100x100 or 1000x1000 grids) and the resulting memory/computational requirements for experience replay buffers and Q-value approximation remains uncharacterized.

  Optimized Autonomous Drone Navigation Using Deep Q-Network Based Reinforcement Learning · 2026 · DOI

• The reward function and ϵ-greedy exploration decay schedule are mentioned but specific numerical values and functional forms are not detailed. The sensitivity of DQN convergence and path optimality to reward shaping, discount factor γ, learning rate, and exploration decay rates in drone navigation scenarios requires systematic ablation analysis.

  Optimized Autonomous Drone Navigation Using Deep Q-Network Based Reinforcement Learning · 2026 · DOI

• The DQN architecture uses a fixed two-hidden-layer fully connected network with ReLU activation. The comparative performance of alternative architectures (convolutional neural networks for image-based state inputs, recurrent networks for temporal dependencies, dueling DQN variants) for drone navigation in environments with partial observability has not been evaluated.

  Optimized Autonomous Drone Navigation Using Deep Q-Network Based Reinforcement Learning · 2026 · DOI

• All experiments were conducted on a non-hardware-accelerated CPU platform in Python. The computational feasibility and inference latency of the trained DQN model on embedded drone hardware (e.g., NVIDIA Jetson, ARM-based flight controllers) for real-time autonomous navigation has not been validated.

  Optimized Autonomous Drone Navigation Using Deep Q-Network Based Reinforcement Learning · 2026 · DOI

• The state representation uses Manhattan distance to the goal (dt) and binary obstacle presence (ot), but the impact of more sophisticated obstacle proximity encoding (e.g., multi-range distance sensors, occupancy grid representations) on DQN convergence speed and path safety in cluttered environments remains unexplored.

  Optimized Autonomous Drone Navigation Using Deep Q-Network Based Reinforcement Learning · 2026 · DOI

• The DQN-based drone navigation framework was evaluated exclusively in a discrete, two-dimensional grid environment with predetermined obstacle layouts. The generalization of this approach to continuous three-dimensional airspace with dynamic obstacles, wind disturbances, and real-time sensor noise has not been investigated.

  Optimized Autonomous Drone Navigation Using Deep Q-Network Based Reinforcement Learning · 2026 · DOI

• Future research should explore the application of alternative classifier induction methods beyond Fuzzy Decision Trees for BDD construction from uncertain data.

  Availability of drone mission with binary decision diagram based on uncertain data · 2026 · DOI

• This paper addressed the key limitations of existing IDSs in the IoV. These limitations include poor detection of diverse DDoS attack types and weak adaptability to high network density, dynamic topology, and limited computational resources. Such issues reduce IDS effectiveness in real-time vehicular environments. To overcome these challenges, an ensemble learning– based IDS was proposed. The framework integrates two optimized CNN models for both binary and multi-class classification. The system effectively detects twelve types of DDoS attacks and adapts to changing IoV conditions. A dynamic deployment algorithm was also introduced to select the best IDS location among fog nodes, RSUs, and UAVs. This ensures continuous detection during congestion and partial system failures. In addition, a game-theoretic optimization strategy was designed to activate the IDS only when an attack is likely. This approach reduces energy con- sumption and limits unnecessary processing overhead. The proposed system was evaluated using three bench- mark datasets: VDoS-LRS, CICDDoS2019, and VDDD. The experimental results showed detection accuracy of over 99%. The proposed method outperformed existing approaches in terms of accuracy, scalability, adaptability, and efficiency. The simulation results also confirmed the effectiveness of the game-theoretic mechanism in limiting attacker advantage. For future work, advanced feature selection techniques such as information gain and Fast Correlation-Based Filter (FCBF) will be investigated. More efficient image-based packet representation methods will also be explored. These improvements aim to reduce per-packet classification time and enhance real-time performance in large-scale IoV systems. Authors’ contributions S.H. and Z.J. designed the study. Z.J. and S.S. and A.M. implemented the proposed model and conducted the experi- ments. S.H. and Z.J. analyzed the results and prepared figures. Z.J. and S.S. and A.M. wrote the main manuscript text. All authors reviewed and approved the final manuscript. Funding The authors declare that no funds, grants, or other support were received for conducting this study. Data availability No datasets were generated or analysed during the current study.

  Enhancing security in iov: an ensemble learning approach for DDoS detection · 2026 · DOI

• The most interesting open questions are probably in AI-driven path planning, solar-assisted charging at delivery hubs, and coordinating multi-drone fleets to share loads and balance battery cycles across a network.

  Smart Energy Optimization for Autonomous Delivery Drones : Insights for E-Commerce Applications · 2026 · DOI

• The latest research trends and open challenges in the field are highlighted, and promising directions for future investigations are identified. Debbah, "A Tutorial on UAVs for Wireless Networks: Applications, Challenges, and Open Problems," IEEE Communications Surveys & Tutorials, vol. Ben Letaifa, "UAV communications with machine learning: challenges, applications and open issues," Arabian Journal for Science and Engineering, vol. Bouaziz, "Federated learning for UAVs-enabled wireless networks: Use cases, challenges, and open problems," vol. , "Internet of drones security and privacy issues: Taxonomy and open challenges," IEEE Access, vol.

  A Comprehensive Survey on 5G-and-Beyond Networks With UAVs: Applications, Emerging Technologies, Regulatory Aspects, Research Trends and Challenges · 2024 · DOI

• (cid:34) (cid:34) (cid:37) (cid:34) (cid:37) (cid:34) (cid:34) (cid:34) (cid:34) (cid:34) (cid:34) (cid:34) (cid:34) (cid:34) (cid:34) (cid:34) (cid:37) (cid:34) operations a sole UAV can effectively execute. Single UAVs encounter challenges in completing missions when faced with rapid battery depletion, extended mission duration, potential electronic system failures due to external or internal factors, or susceptibility to targeting by attackers. These factors significantly hinder the effectiveness of single UAV operations. In these scenarios, FANETs are recommended, as they allow multiple UAVs to join a common network and execute complex tasks in an organized manner. Although FANETs inherits certain features from MANETs and its sub-classes, it also presents differences due to the very characteristics of UAVs such as their high mobility, unpredictable movements, and frequently changing network topology. Subsequently, such characteristics of UAVs are detailed along with the relevant security perspective. A. Node Mobility & Dynamic Topology FANETs differ from other ad hoc networks due to UAVs’ exceptional node mobility. These networks possess highly dynamic topology due to frequent changes in node positions. Mobility models differ in FANETs according to its application. UAVs might follow predetermined paths or move randomly. They might exhibit independent movement or move collectively in group-based models. Unlike nodes in MANETs and VANETs, they maneuver in 3D space. Security Impacts: The highly dynamic nature of the network topology poses a significant challenge in differentiating between normal and abnormal behaviour. For instance, identifying a node that is sending routing misinformation becomes intricate, as it could be an attacker or simply outdated. Furthermore, creating secure systems within dynamically changing environments poses significant architectural challenges. Moreover, high-level mobility can impact security in both positive and negative ways. The mobility of targets, on one hand, can serve to mitigate the impact of attacks directed towards them. Conversely, the mobility also enables attackers to easily evade security measures. B. Node Density Node density, which refers to the average number of UAVs per unit area, can vary from low to high, depending on factors such as operational areas, airspace coverage, applications, and the types of UAVs deployed. If UAVs possess high speeds and wide transmission ranges, their density tends to diminish as the distances separating them could extend across several kilometers. Consequently, node density in FANETs is typically observed to be lower compared to both MANETs and VANETs. Security Impacts: In scenarios with high node density, certain attacks like sinkholes can be particularly effective due to the increased connectivity. Such attacks exploit the density by attracting and redirecting network traffic, posing significant security risks.

  A Survey of Security in UAVs and FANETs: Issues, Threats, Analysis of Attacks, and Solutions · 2024 · DOI

Most-cited papers in UAV Applications and Optimization

• Multi-Agent Reinforcement Learning-Based Resource Allocation for UAV Networks · IEEE Transactions on Wireless Communications · 2019 · 525 citations
• Unmanned Aerial Vehicles in Smart Agriculture: Applications, Requirements, and Challenges · IEEE Sensors Journal · 2021 · 507 citations
• Multi-Agent Deep Reinforcement Learning for Task Offloading in UAV-Assisted Mobile Edge Computing · IEEE Transactions on Wireless Communications · 2022 · 417 citations
• Coordinated Logistics with a Truck and a Drone · Management Science · 2017 · 359 citations
• Unmanned Aerial Vehicles for Search and Rescue: A Survey · Remote Sensing · 2023 · 343 citations
• Deep Reinforcement Learning for Delay-Oriented IoT Task Scheduling in SAGIN · IEEE Transactions on Wireless Communications · 2020 · 326 citations
• A Comprehensive Review of Unmanned Aerial Vehicle Attacks and Neutralization Techniques · Ad Hoc Networks · 2020 · 325 citations
• 3D UAV Trajectory Design and Frequency Band Allocation for Energy-Efficient and Fair Communication: A Deep Reinforcement Learning Approach · IEEE Transactions on Wireless Communications · 2020 · 297 citations
• Joint Maneuver and Beamforming Design for UAV-Enabled Integrated Sensing and Communication · IEEE Transactions on Wireless Communications · 2022 · 293 citations
• Deep Reinforcement Learning Based Dynamic Trajectory Control for UAV-Assisted Mobile Edge Computing · IEEE Transactions on Mobile Computing · 2021 · 284 citations

Most recent work

• AirFogSim: A Light-Weight and Modular Simulator for UAV-Integrated Vehicular Fog Computing · IEEE Transactions on Mobile Computing · 2026
• AeroResQ: Edge-accelerated UAV framework for scalable, resilient and collaborative escape route planning in wildfire scenarios · Future Generation Computer Systems · 2026
• Availability of drone mission with binary decision diagram based on uncertain data · Scientific Reports · 2026
• UAV swarm safe coverage path planning with deep reinforcement learning · Discover Computing · 2026
• Smart Energy Optimization for Autonomous Delivery Drones : Insights for E-Commerce Applications · International Journal of Advanced Research in Science Communication and Technology · 2026
• Enhancing security in iov: an ensemble learning approach for DDoS detection · Peer-to-Peer Networking and Applications · 2026
• Optimized Autonomous Drone Navigation Using Deep Q-Network Based Reinforcement Learning · International Research Journal on Advanced Engineering and Management (IRJAEM) · 2026
• Optimizing Clustering through ACO and Secure Waterfall Energy-Efficient Protocol-Enabled Routing in FANETs · INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT · 2026
• Design of Sky Watcher Intelligent Drone Detection System · International Scientific Journal of Engineering and Management · 2026
• Harsh Weather-Oriented Edge Intelligence Empowered Maritime Communication-Computing Converged Network Resource Allocation · IEEE Internet of Things Journal · 2026

Related topics in Engineering