Resource
2026 EN
Yoshihide Aoyanagi · Hajime Arai · Tomofumi Doi
+2 more
The number of microsatellites and CubeSats in low-Earth orbit has rapidly increased, creating a strong need for autonomous mission operations that can function without continuous ground intervention. In particular, Earth observation CubeSats often suspend operations when battery voltage becomes low, making power-aware autonomous planning essential for maintaining stable mission execution. This study proposes an onboard future-state prediction and autonomous observation system that allows CubeSats to schedule and execute observations while maintaining a sufficient power margin. The system integrates a seasonal autoregressive integrated moving average (SARIMA) model for battery-voltage forecasting, an onboard battery state-of-charge estimation scheme, and an autonomous observation planner. These functions were implemented on a compact edge-computing module designed for resource-limited spacecraft. The developed module was evaluated through radiation testing, confirming its suitability for on-orbit operation. Ground-based validation confirmed that the proposed method can accurately reproduce the medium-term behavior of the battery’s charge–discharge cycle based on housekeeping telemetry. The system was implemented on a CubeSat and subsequently demonstrated during orbital operations. During autonomous operation, the satellite predicted its future power state and successfully completed all planned observations using its onboard camera. The predicted battery voltage showed strong agreement with the measured trend, providing sufficient reliability for onboard decision-making. These results demonstrate that lightweight time-series forecasting combined with autonomous planning enables power-aware and self-managed operations on resource-constrained small satellites, establishing a foundation for future fully autonomous missions.
Resource
2026 EN
M. Damircheli · M. Fakoor · P. Nikpey
This study presents the Controller Reliability Assessment Model (CRAM), a new framework for evaluating the reliability of the controller and the attitude control subsystem (ACS) of Low Earth Orbit (LEO) satellites. CRAM integrates Software-in-the-Loop (SIL) simulations to construct a performance-based database that captures the operational behavior of the ACS under both internal and external disturbances. Data-driven analysis is then applied to derive a detailed understanding of controller performance and to determine the failure rates of controller components. In addition, a Reliability Block Diagram (RBD) is developed to quantify the contribution of each controller and to estimate system-level reliability over a three-year mission duration. The proposed model identifies the most robust, accurate, and responsive controller with the highest reliability. The ACS design incorporates solar radiation pressure, gravity-gradient torque, and aerodynamic drag as external disturbances, while actuator loss of effectiveness is modeled as an internal disturbance. A tetrahedral reaction wheel configuration is employed, where three wheels provide three-axis control and a fourth wheel serves as redundancy. The evaluated control schemes include a proportional–derivative (PD) controller, an H∞ controller, and two variants of sliding-mode control.
Resource
2026 EN
Nuno A. D. Soares · Antonio M. R. C. Grilo
The Resource-Constrained Shortest Path Problem (RCSPP) is a fundamental NP-hard optimization challenge with broad applications ranging from network routing and logistics to autonomous navigation. Although a rich body of exact RCSPP solvers exists, their practical applicability is severely limited on large-scale, dense graphs, where classical label-setting methods suffer from exponential growth in the number of non-dominated labels. This limitation is particularly critical for Unmanned Ground Vehicle (UGV) mission planning, where route generation is performed prior to deployment using high-resolution terrain representations and heterogeneous threat models that naturally induce dense grid-based graphs and strict travel-time budgets. This paper introduces APULSE, a hybrid search algorithm designed to address this scalability gap by combining A* heuristic guidance, aggressive Pulse-style pruning, and a time-bucketing mechanism that bounds memory consumption through a controlled relaxation of the dominance relation. As a consequence of this discretization, APULSE is not guaranteed to return the exact RCSPP optimum for all instances; instead, it offers a tunable accuracy–efficiency trade-off governed by an auto-tuning parameter that adjusts temporal granularity. A comprehensive computational study evaluates APULSE against state-of-the-art exact solvers and Beam Search heuristics on a large, dense UGV planning graph. Under the default auto-tuning configuration, APULSE matches the empirical reference solution returned by exact solvers in 8 out of 9 evaluated scale and slack configurations, with a maximum observed optimality gap of 0.01% in the remaining case, and empirically converges to the reference solution as the bucketing resolution is increased. At the same time, APULSE remains orders of magnitude faster and substantially more memory-efficient than exact methods on the largest dense instances. These results demonstrate that APULSE enables practical constrained route generation and rapid what-if analysis on dense graphs within operationally realistic decision-making windows.
Resource
2026 EN
Ayman Elshenawy · Khalil M. Abdelnaby · Abdurrahman A. Nasr
+1 more
The proliferation of Hardware Trojans (HTs) poses a critical threat to the reliability and security of integrated circuits (ICs), particularly in safety and mission-critical domains. Conventional detection approaches either rely on destructive imaging or trusted golden chips, both of which are costly and limited in scalability. This study introduces a golden-chip-free, non-destructive framework for Hardware Trojan Detection (HTD) based on transformer architecture applied to Power Side-Channel (PSC) signals. The research contribution presents the first systematic performance comparison between decoder-only (GPT-like) and encoder-only (BERT-like) models, as well as full encoder–decoder transformer models, for time-series PSC analysis. To adapt these models to the hardware security domain, side-channel traces are normalized, reduced via principal component analysis, and embedded with positional encodings to preserve temporal structure. The framework is evaluated on the publicly available Trojan Power & EM Side-Channel dataset using 10-fold cross-validation. Results show that GPT-based and BERT-based models achieve 66.32% and 79.41% accuracy, respectively. While the full transformer network significantly outperforms them with 92.43% accuracy, 92.44% precision, 92.43% recall, and an F1-score of 92.31%. These findings demonstrate the potential of attention-based architectures to capture subtle variations in PSC signals without requiring a golden reference chip. The study contributes a reproducible and comparative benchmark for transformer-based HTD, offering a scalable pathway toward more robust and interpretable hardware security solutions.
Resource
2026 EN
Ahmad Merei · Hamid Mcheick · Alia Ghaddar
+1 more
Uncrewed Aerial Vehicles (UAVs) have become increasingly vital in various missions due to their ability to access remote and hazardous environments. Effective deployment of UAVs in time-sensitive and energy-constrained missions requires effective path planning. These strategies must maximize mission utility while operating within strict energy and time budgets. This paper introduces an Energy-Aware Informative Path Planning algorithm (EA-IPP2n) that functions as a pre-mission planner, generating the UAV’s flight path prior to deployment. The algorithm simultaneously accounts for both distance and turning angle costs. This enables the UAVs to plan efficient paths in terms of energy consumption and completion time in reward-driven environments. The method models the environment as a graph of nodes that represent locations of interest with associated rewards. EA-IPP2n employs a two-hop look-ahead mechanism, in which the planner evaluates pairs of consecutive moves before committing to the next step, to prioritize high-reward while accounting for both budgets. Simulations show improved performance over baseline methods, including Hierarchical Informative Path Planning (HIPP), Multiple-UAV Informative Path Planning and Mapping (MIPP), Ant Colony Optimization (ACO), Genetic Algorithm (GA), Monte Carlo Tree Search (MCTS), and Greedy Search. EA-IPP2n achieved the highest total rewards (24.9 ± 3.6), exceeding the baseline algorithms by relative margins between 11% and 66%, while fully satisfying both time and energy constraints across all missions. It also achieves the highest mission performance score (0.380 ± 0.041), with an improvement of approximately 11% compared to the best-performing baseline, and relative gains between 11% and 57% across all evaluated algorithms. Furthermore, EA-IPP2n strictly respected both budgets in 100% of missions. While the time budget was enforced for all algorithms, only the enforced dual-budget formulation enabled HIPP and MIPP to satisfy both constraints in all cases. Greedy and GA violated the energy budget in all missions, and MCTS exceeded its limit in 40% of missions. In contrast, ACO, despite being heuristic and less informed, maintained full energy compliance. Thus, EA-IPP2n was the only method that achieved complete dual-budget compliance while maintaining the highest total reward performance.
Resource
2026 EN
Seong Min Kim · Jun Lee · Hyun Kwon
This paper presents an integrated AI-driven framework that unifies semantic segmentation and deep reinforcement learning (DRL) to enable tactical pathfinding in battlefield environments characterized by complex terrain and adversarial threats. High-resolution aerial imagery captured by reconnaissance drones is processed with a lightweight yet accurate segmentation model to classify terrain into eleven operational categories, such as roads, forests, rice paddies, and restricted zones. The resulting semantic map is transformed into a cost-weighted potential field, where each terrain type is assigned a traversal weight reflecting its impact on mobility and mission risk. A Deep Q-Network (DQN) agent is then trained to optimize its navigation policy by selecting appropriate repulsive potential coefficients that minimize environment-weighted travel time while adaptively avoiding impassable regions and predicted enemy zones. Through continuous interaction with the environment, the agent adaptively adjusts its repulsive potential coefficient to ensure both efficiency and safety. Experimental results demonstrate that the proposed framework outperforms conventional pathfinding approaches in terms of path efficiency and threat avoidance, offering a practical and scalable solution for real-time decision-making in autonomous military mobility.
Resource
2026 EN
Taewan Cho · Hyunwoo You · Andrew Jaeyong Choi
The future of disaster response, search and rescue (SAR) technology is rapidly evolving with the integration of Artificial Intelligence (AI) and intelligent unmanned platforms. Manned-Unmanned Teaming (MUM-T) is revolutionizing on-site disaster structure and emergency rescue strategies by synchronizing manned and unmanned vehicles to support operations. This paper investigates the potential applications of Multimodal Large Language Models (MLLMs) in intelligent MUM-T systems. We present a system capable of using aerial imagery to identify and locate survivors and hazards/risks, and to generate safe path plans for unmanned ground vehicles (UGVs). The main contributions are: (1) the implementation of a Few-Shot In-Context Learning (ICL) method for MLLMs, which eliminates the need for fine-tuning or retraining the AI model, and (2) the application of the Chain-of-Thought (CoT) prompting technique. The ICL method reduces the time required for data gathering and model training, while the CoT reasoning technique improves the path planning generated by MLLMs. Our experiments, using aerial imagery, demonstrate that the proposed system achieves up to 86% mean Average Precision (mAP) in survivor and hazard identification through 5-shot ICL. Furthermore, applying the CoT technique doubled the UGV’s path generation success rate from 20% to 40%. The proposed MLLM-based system effectively performed open-vocabulary object detection without additional training. This research highlights the potential of MLLMs for diverse and complex disaster relief missions in intelligent MUM-T systems, presenting an innovative approach to enhance capabilities and mission success while reducing human risk and operational costs. A demonstration video of our system is available at https://youtu.be/vReaSMDWu8o?si=RluQsqS4O_j5lVcS.
Resource
2026 EN
Kunying Wang · Hengli Jin · Ronghuan Yu
The proliferation of large-scale Earth observation satellite constellations has substantially increased mission scheduling complexity, where dynamic operational constraints, heterogeneous task requirements, and stringent real-time demands must be handled concurrently. Deep reinforcement learning has emerged as a promising paradigm for autonomous scheduling by enabling agents to learn adaptive policies through interaction with simulated mission environments. This article presents a systematic review of deep reinforcement learning approaches to Earth observation satellite scheduling, synthesizing 54 eligible studies identified through a PRISMA-style screening workflow spanning IEEE Xplore, Web of Science, Scopus, and a supplementary Google Scholar search.We introduce a reference constrained Markov decision process formulation that consolidates notation and modeling choices across the reviewed literature, serving as a structured analytical lens for cross-study comparison rather than a novel algorithmic contribution. Building on this formulation, we categorize representative methods into value-based, policy-based, and model-based families, and provide critical assessments of convergence behavior, scalability, constraint satisfaction, and deployment readiness for each paradigm. A structured cross-study synthesis maps 30 representative studies against seven modeling dimensions, revealing field-wide patterns and evidence gaps. The review concludes with a standardized minimum reporting checklist to improve reproducibility and with concrete research directions toward flight-qualified autonomous scheduling.
Resource
2026 EN
Hamad Alblooshi · Kevser Ovaz Akpinar · Mustafa Akpinar
The Industrial Internet of Things (IIoT) is accelerating digital transformation in industrial sectors through connected sensors, edge devices, and data-driven automation; however, this connectivity also increases exposure to cyber threats that can disrupt safety-critical and mission-critical operations. This study addresses the need for effective IIoT intrusion detection by proposing a hybrid model that combines a genetic algorithm (GA) with a neural network (NN), in which the GA optimizes NN parameters to improve classification performance in complex traffic environments. The proposed approach is evaluated on the Edge-IIoTset dataset, a labeled IIoT intrusion detection benchmark that includes normal traffic and multiple attack families (e.g., DDoS, malware, and injection-related attacks), represented by heterogeneous feature types and characterized by pronounced class imbalance. Dimensionality reduction is applied using principal component analysis to reduce feature redundancy and improve learning efficiency. Experimental results show that the GA-optimized NN achieves up to 95% overall accuracy across the evaluated attack categories, with particularly strong discrimination for dominant attack groups. The findings indicate that evolutionary optimization can enhance the performance of NN-based IDSs and provide a flexible optimization framework for IIoT settings, thereby supporting the development of more resilient intrusion detection solutions for industrial networks.
Resource
2026 EN
Luu Trong Nhan · Timote Moreaux · Victor Khaustov
+8 more
As the global energy sector pivots toward "gigantic" offshore wind structures, the logistical and safety challenges of marine construction have reached a critical bottleneck. This paper pioneers a transformative reinforcement learning (RL) framework designed to automate offshore cement wind turbine construction through a specialized cluster of high-performance quadcopter agents. To eliminate the optimization interference and "negative transfer" typical of multi-task systems, we introduce a modular architecture that decomposes the construction workflow into distinct, high-precision phases: autonomous navigation, cement dispensing, and energy-aware refueling. At the core of our system is the novel application of Joint Spectral Normalization (SN), which stabilizes training by bounding the Lipschitz constants of deep networks, ensuring monotonic policy improvement even in volatile environments. Benchmarking against five state-of-the-art algorithms (SAC, DDPG, TD3, TRPO, and PPO) reveals that our PPO-based SN framework is the only solution capable of consistent convergence. Crucially, while vanilla RL controllers collapse under stochastic disturbances, our Full SN variant maintains mission-critical stability and high convergence rates across extreme Gaussian and uniform observation noise. This study marks the first successful integration of RL-driven aerial robotics into complex industrial construction, offering a robust path forward for civil engineering automation in the world’s most unpredictable marine settings.