Search Results (1,929)

This study proposes a surrogate model, Graph-Structured Physics-Informed Deep Operator Network (GS-PI-DeepONet), to simulate two-dimensional hydrodynamics in a tidal river network. The model was coupled with Graph Convolutional Networks (GCNs) for spatial feature extraction and Long Short-Term Memory (LSTM) network for temporal prediction, with 2D shallow-water equations (2D SWEs) embedded as physical constraints. To handle complex river network topologies, a mapping mechanism was proposed to transform discrete irregular boundaries into differentiable neural network constraints. A dynamic weighting strategy was developed to improve model training efficiency. GS-PI-DeepONet was applied to a river network within the Pearl River Basin in Zhuhai. Trained on high-fidelity Delft3D data, it achieved precise flow field reconstruction and millisecond-level extrapolation predictions, outperforming traditional data-driven models. The model can be a valuable tool for real-time hydrodynamic simulations and flood management strategies in tidal river networks. Full article

Image compression has evolved from human-centric perceptual coding toward support for diverse machine vision applications, requiring modern codecs to serve both human viewing and downstream tasks in closed-set settings (where target tasks are incorporated during training) and open-set settings (where previously unseen tasks arise at test time). While recent learned compression methods jointly optimize perceptual quality and closed-set task performance, they often fail to generalize to unseen open-set tasks due to fixed training assumptions and objectives. Our prior work, LoRA-comp (Low-Rank Adaptation Compression), effectively addresses open-set challenges via instance-specific test-time fine-tuning (TTFT) without requiring task-specific pre-training. Nevertheless, its fixed LoRA architecture, which assigns a uniform rank across all layers, often leads to suboptimal instance-level performance. Moreover, allocating the same number of training epochs to every instance introduces unnecessary encoding-time overhead. To address these challenges, we propose Instance-Specific Early Stopping with Dynamic Rank Adaptation (ISER), which extends LoRA-comp. Building upon the LoRA-comp–based instance-specific adaptation framework, ISER introduces (i) instance-specific early stopping (ISES) combined with a multi-scale training strategy (MSTS) to reduce TTFT overhead and (ii) instance-specific dynamic rank adaptation (ISRA) to tailor the LoRA architecture per instance. Experiments demonstrate that ISER consistently outperforms competing methods. Compared to LoRA-comp, ISER achieves up to a 7% BD-Rate improvement and up to a 44% reduction in encoding time. Moreover, ISER achieves up to a 98% relative improvement in BD-Rate gain and up to a 19.2% reduction in decoding time over TransTIC. Full article

Fast charging of lithium-ion batteries requires balancing charging efficiency with electrochemical safety to minimize degradation and lithium plating. Conventional charging strategies and existing reinforcement learning approaches often lack physical consistency and model interpretability, limiting their applicability in safety-critical battery management systems. This paper proposes a physics-informed Attention-Proximal Policy Optimization (Attention-PPO) framework for intelligent battery fast charging by integrating the Single Particle Model with Electrolyte (SPMe) with a transformer-based attention mechanism. SPMe provides physically meaningful battery state transitions, while the attention-enhanced PPO dynamically learns informative electrochemical representations for charging control. To improve transparency, a monotonic XGBoost surrogate model is employed for lithium-plating risk estimation, and the learned policy is further distilled into an interpretable decision tree. Experimental results demonstrate that the proposed Attention-PPO achieves substantially faster and more stable policy convergence than the baseline PPO, reaching convergence at episode 37 compared with episode 82 for PPO, corresponding to a 54.9% reduction in training episodes. The reward standard deviation is also reduced from 0.132 to 0.041, indicating 68.9% lower reward variability. In terms of electrochemical safety, Attention-PPO achieves a mean plating overpotential of +0.023 V compared with −0.015 V for PPO, providing a 38 mV improvement and a positive safety margin against lithium plating. Compared with conventional CC-CV and CC-COP controllers, the proposed framework requires a longer charging duration because it prioritizes electrochemical safety; however, it consistently maintains positive plating overpotential while achieving reliable charging performance. Compared with conventional CC-CV, CC-COP, and standard PPO controllers, the proposed framework provides a larger electrochemical safety margin while preserving reliable charging performance. Transformer attention analysis and policy distillation provide interpretable representations of the learned charging policy, while SHAP analysis characterizes feature contributions within the auxiliary plating-risk estimator. The proposed framework provides an effective and explainable physics-informed reinforcement learning solution for safe lithium-ion battery fast charging, offering a promising approach for next-generation intelligent battery management systems. Full article

Football match outcomes are influenced by a complex interplay of dynamic team strategies and stochastic match events, posing a significant challenge for predictive analytics. In this paper, we present a unified benchmarking study for football match outcome classification, using a dataset of 225,474 matches spanning 2018–2026 and a strictly chronological train–test split. Six classifiers spanning several modelling approaches, Logistic Regression, Generalized Additive Models, FastTree, Random Forest, Radial Basis Function networks and Multi-Layer Perceptron, are trained and evaluated under identical conditions. Bookmaker odds are converted into margin-free probabilities by the power method, with the exponent found by the Newton–Raphson method, and are included among the input features. In addition to the six classifiers, we also report a majority-class baseline and the de-vigged bookmaker prediction itself. Results are reported for both a binary one-vs-rest formulation and the original three-class (1, X, 2) formulation, with 95% bootstrap CIs for every metric and every model. Across the six classifiers, the differences are small: macro-averaged Precision ranges from 64.83% to 65.16% for Home Win and from 68.05% to 68.46% for Away Win, and macro-averaged Recall from 63.16% to 63.51% and from 59.10% to 59.72%, respectively, with substantially overlapping confidence intervals. In the three-class formulation, models trained without market data reach 48.0% accuracy against 43.4% for a trivial baseline and 51.9% for the bookmaker. Models trained with market data match the bookmaker but do not improve upon it on log-loss, Brier score or the ranked probability score. The only statistically distinguishable improvement of any kind is a 0.086 point accuracy advantage for the additive model, which is not accompanied by any improvement in the proper scoring rules and therefore does not indicate a practical advantage. Probability calibration and betting-signal generation are outside the scope of this study. Full article

Hyperspectral image super-resolution (HSI SR) focuses on enhancing the spatial resolution of HSIs while preserving their inherent spectral information. Existing single-image HSI SR methods still suffer from blurred spatial edges and spectral distortion. Although numerous spatial–spectral enhancement networks can enhance spatial–spectral feature extraction, they often lead to a cumbersome network architecture. To address these issues, we propose a Spatial–Spectral Distillation Mamba Network, called SSDM-Net, for HSI SR, which contains a main reconstruction branch and two training-only auxiliary branches for spatial and spectral knowledge distillation. Specifically, the spatial and spectral auxiliary branches, which are utilized exclusively during training, provide edge-aware guidance and capture spectral correlations, respectively. During training, the spatial–spectral knowledge is transferred to the main branch. During inference, the auxiliary branches are removed, improving reconstruction quality without extra computational burden. In the main branch, a Mamba-based spatial–spectral global enhancement module processes spatial and latent inter-channel sequences using selective scanning whose cost is linear in the processed sequence lengths when the feature dimensions are fixed. In addition, a dynamic loss weighting strategy is developed to balance reconstruction, distillation, and auxiliary losses during optimization. Comprehensive experiments conducted on the CAVE and Houston datasets with three scale factors demonstrate that SSDM-Net produces more accurate reconstruction results than existing representative HSI SR methods. Cross-dataset experiments on the Harvard dataset further suggest that the method can maintain competitive reconstruction performance under the evaluated cross-dataset settings. Full article

Caspase-1 is a crucial inflammatory cysteine protease that facilitates the maturation of pro-inflammatory cytokines such as interleukin-1β and interleukin-18, making it a significant therapeutic target for inflammatory diseases. However, existing caspase-1 inhibitors often face challenges like toxicity and suboptimal drug-like properties, underscoring the need for new inhibitors. This study employed an integrated computational strategy, combining quantitative structure–activity relationship (QSAR) modeling and application of this validated model to a large natural product database followed by molecular docking, rigorous binding free energy analysis and extended molecular dynamics simulations. Initially, a dataset of 185 caspase-1 inhibitors with experimentally reported pKi values (ranging from 4.05 to 9.24) was used to construct a QSAR model using Partial Least Squares (PLS) regression. The PLS-based QSAR model was developed with 18 descriptors out of 5799 calculated descriptors for each compound and 10 latent variables, demonstrating strong statistical performance with R² values of 0.870 and 0.838 for the training and test sets, respectively, and leave-one-out cross-validation coefficient Q²_LOO and 5-fold cross-validation (Q²_5-fold) values of 0.819 and 0.814, respectively. Y-randomization tests further confirmed the model’s robustness, as the randomized models exhibited significantly lower statistical parameters than the original model. The validated QSAR model was applied to 276,518 natural products in the LOTUS database. Subsequent molecular docking, Molecular Mechanics/General Born Surface Area (MM/GBSA) scoring, and Pan-Assay INterference Compounds (PAINS) and Chemical Frequent Hitter (ChemFH) filtering identified 14 candidate compounds, which were further evaluated using 300 ns molecular dynamics simulations. Among these, four natural products (LTS0162325, LTS0221286, LTS0016840, and LTS0070407) showed the most stable binding behavior and maintained persistent interactions with key catalytic and substrate-binding residues of caspase-1 in a mimicked physiological condition. Overall, this study highlights natural diterpenoids and coumarin glycosides as promising scaffolds for caspase-1 inhibition and demonstrates that integrating QSAR modeling with structure-based approaches provides an efficient strategy for discovering potential anti-inflammatory drug candidates. Full article

Large-scale maritime activity, persistent shipping incidents, and complex marine environments continue to place substantial demands on maritime search and rescue (MSAR). Current MSAR systems do not fully capitalize on the complementary strengths of unmanned aerial vehicles (UAVs) and satellites for collaborative tracking and rescue support. Existing air-space collaboration technologies suffer from two critical limitations: (1) rigid processes, including fixed task allocation, pre-determined path planning without real-time environmental adaptation, and isolated satellite–UAV decision-making, and (2) long task completion cycles, mainly because many methods are adapted to wide-area, long-duration military tracking scenarios. They therefore provide limited support for the dynamic flexibility required in MSAR. This study proposes a Proximal Policy Optimization (PPO)-based air-space collaborative tracking method for maritime moving targets to address these shortcomings and enhance air-space cooperation in MSAR operations. The core implementation of the method includes: (1) integration of target drift forecasting, satellite orbit prediction, UAV task allocation, and path planning into a unified reinforcement learning framework to reduce isolated single-platform decision-making; (2) the adoption of PPO to generate dynamic and flexible air-space collaborative tracking strategies that adjust satellite observation angles and scanning ranges, as well as UAV altitude, speed, and heading according to real-time target, environmental, and platform states; and (3) the design of a multi-dimensional reward function that balances target proximity, energy efficiency, coverage overlap, and inter-platform cooperation to guide strategy optimization. Simulation experiments include system-feasibility verification, baseline-controller comparison, PPO hyperparameter screening, and cross-scenario evaluation. Under idealized communication and payload-matching assumptions, the method enables coordinated tracking of maritime moving targets in simulated MSAR scenarios. In the standardized evaluation, PPO achieved an 11.9% higher mean evaluation episode return, 11.2% lower aggregate UAV energy consumption, and a 9.92-percentage-point greater endurance margin than DDPG. Hyperparameter screening compared candidate learning rates, discount factors, and training budgets, informing the PPO configuration for the subsequent six-scenario evaluation. Across the six controlled scenarios, rewards stabilized after approximately 1400 steps, while action magnitudes varied among regions. These results indicate that the proposed method has potential to enhance air-space collaborative tracking for MSAR decision support. Full article

Ambient Assisted Living (AAL) environments aim to enable older people to remain active and lead an autonomous and independent life for as long as possible. Among the technologies that can be incorporated into these settings, socially assistive robots (SAR) seek to establish a more natural and intuitive means of interaction with people, whilst helping them to carry out everyday tasks. One of the main challenges facing the design of these robots is how to enable them to undertake more complex tasks. Recent advances in Large Language Models (LLMs) have opened new avenues for flexible robot deliberation, yet their integration into real-time robotic systems remains challenging due to latency constraints, reasoning reliability, and the complexity of coordinating multi-step tasks. This paper proposes a hierarchical multi-agent architecture for robot deliberation that addresses these challenges by combining LLM-based planning with structured execution mechanisms within the ROS 2 ecosystem. The proposed architecture employs a supervisor agent that decomposes high-level natural language instructions into prioritised subtasks, enabling a priority-driven execution model that dynamically adapts to task relevance, temporal constraints, and environmental feedback. Subtasks are delegated to a set of Single-Purpose Agents (SPAs), orchestrated via LangGraph state machines and coordinated through a priority-aware scheduling mechanism. A key design principle is the use of Behaviour Trees (BTs) as high-level callable tools through the Model Context Protocol (MCP), encapsulating closed-loop control strategies while enabling preemptive and priority-consistent execution. This reduces the number of LLM inference steps required per task and improves robustness under dynamic conditions. A further contribution concerns the deployment of fine-tuned, lightweight LLMs—on the order of 0.6 billion parameters—specifically adapted for both the supervisor and the individual SPA roles through parameter-efficient low-rank adaptation (LoRA). These models are trained on role-specific tool-calling datasets to specialise in constrained reasoning patterns and task-specific decision-making, enabling efficient, low-latency inference directly on edge hardware. The combination of fine-tuning and hierarchical priority control enhances both the determinism and responsiveness of the system while mitigating error propagation across agent interactions. The paper presents the full software architecture, a formal characterisation of the system as a priority-aware hierarchical policy over a graph of agent workflows, and an experimental evaluation in an Ambient Assisted Living scenario assessing task success rate, inference efficiency, responsiveness under competing priorities, and overall user experience. Because SPA execution is decoupled from the supervisor’s own reasoning loop, the architecture is designed to keep accepting, processing, and queuing new user queries while previously dispatched SPAs are still executing their tasks. Full article

Peripheral vestibulopathy (PV) is a common disorder that causes dizziness and balance impairment, substantially affecting patients’ quality of life. When symptoms persist, they frequently lead to anxiety, depression, and an increased risk of social isolation. Although central vestibular compensation (CVC) often promotes functional recovery, current pharmacological options remain limited and are primarily aimed at symptom control. Consequently, they should be considered adjuncts rather than alternatives to rehabilitation-based strategies. Vestibular rehabilitation (VR) remains the cornerstone of treatment for promoting functional recovery in patients with PV. Based on the complementary mechanisms of adaptation, substitution, and habituation, VR enhances the central nervous system’s ability to compensate for vestibular deficits by optimizing the integration of visual, proprioceptive, and residual vestibular inputs. Robust evidence from both clinical and preclinical studies has demonstrated its efficacy in improving postural stability, dynamic balance, gaze stabilization, and overall functional performance. Experimental studies using animal models have further highlighted the critical role of active sensorimotor training in ecologically relevant environments for enhancing vestibular compensation. Pharmacological interventions may further facilitate these adaptive processes by modulating the neurobiological mechanisms underlying vestibular compensation, thereby improving responsiveness to rehabilitation. Likewise, emerging neuromodulation approaches, including galvanic vestibular stimulation, have shown promising potential to enhance neural plasticity and augment the effects of rehabilitation. Consequently, combining VR with targeted pharmacological therapies and/or vestibular stimulation techniques may provide synergistic benefits and maximize functional recovery. In conclusion, vestibular rehabilitation should remain the foundation of PV management and be integrated with pharmacological and neuromodulatory approaches within a multidisciplinary therapeutic framework. Further research is needed to optimize rehabilitation protocols, identify the biological determinants of successful vestibular compensation, and develop personalized therapeutic strategies aimed at maximizing functional recovery and improving patients’ quality of life. Full article

Lower-limb rehabilitation exoskeletons have the potential to improve gait recovery after stroke by providing intensive and repetitive training. However, conventional control strategies often rely on fixed control parameters and exhibit limited adaptability to patient-specific characteristics, sensor noise, and dynamic uncertainties. This paper proposes an adaptive control framework that combines deep reinforcement learning (RL) with model-based impedance control for personalised lower-limb exoskeleton assistance. Patient-specific biological parameters are incorporated into the simulation environment and reward formulation to improve adaptability and robustness. Three state-of-the-art deep RL algorithms, Deep Deterministic Policy Gradient (DDPG), Twin Delayed Deep Deterministic Policy Gradient (TD3), and Soft Actor-Critic (SAC), are evaluated in a continuous control environment under varying signal-to-noise ratio (SNR) conditions ranging from 5 dB to noise-free conditions. Results demonstrate that TD3 achieves the most stable learning performance, obtaining a mean reward of −354.24 under noise-free conditions, while DDPG provides the highest joint-angle tracking accuracy with an RMSE of 0.0369 rad. SAC exhibits superior robustness in noisy environments, achieving the highest learning ratio of 0.51 at 5 dB SNR. Furthermore, the proposed personalised framework reduces tracking errors by up to 27% compared with non-personalised baseline approaches. The findings indicate that integrating patient-specific information with RL-based adaptive control can significantly enhance robustness, tracking performance, and personalisation in exoskeleton-assisted gait rehabilitation, providing a promising direction for future intelligent rehabilitation systems. Full article

Variable illumination, severe occlusion, and the small scale of pest instances in complex forest environments significantly compromise the accuracy of automated forest pest detection. Existing deep learning-based detection models often face challenges in achieving an optimal trade-off among detection accuracy, model lightweightness, and training stability. To achieve efficient and intelligent monitoring of forest pests, we propose a novel detection framework named DAAL-YOLOv12n by improving the lightweight YOLOv12n baseline with two specifically designed modules for complex forest scenarios. To overcome the limited feature adaptability caused by fixed convolutional branches and conventional residual connections in the original A2C2f block, a DynamicA2C2f dynamic attention module is introduced. This module employs dynamic gating weights and adaptive residual adaptation modulation to achieve directional feature enhancement and improved cross-channel compatibility. Furthermore, to reduce the computational redundancy of the conventional C3/C2f block and alleviate abnormal optimization errors in the exponential moving average (EMA) caused by persistently stored dynamic tensors, an AdaptiveC3K2 lightweight adaptation module is developed. Through shared convolution representations, temporary dynamic weight generation, and pre-residual adaptation strategies, the proposed module achieves model compression, flexible channel adaptation, and EMA-compatible optimization. Extensive experiments are conducted on a self-collected forest pest dataset comprising 16 common pest categories. Experimental results demonstrate that, compared with advanced detectors such as YOLOv12n, DAAL-YOLOv12n improves precision and recall by 10.5% and 11.1%, respectively, with only a marginal increase in model parameters. Furthermore, the proposed approach increases mAP@0.5 and mAP@0.5:0.95 by 10.2% and 18.0%, respectively, attaining an mAP@0.5 of 93.0% alongside a real-time inference throughput of 93 FPS. These results indicate that the proposed method offers a viable solution for deploying accurate and lightweight forest pest detection systems on edge devices, offering technical support for intelligent pest surveillance in the wild. Full article

Generalist robots need to perform diverse tasks while operating in dynamic, uncertain, and unstructured environments, often around human beings. Vision-language-action (VLA) models have recently emerged as a promising and flexible framework for integrating perception, reasoning, robotic control, and action execution to develop generalist robotic policies. This systematic literature review (SLR) examines more than 140 VLA-related publications between 2020 and 2025 following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. To the best of our knowledge, it is the first PRISMA-compliant systematic review dedicated to VLA models, offering a structured discussion of robotic policies, VLA architectures, and inference optimization methods. The review also presents descriptive analyses of the included studies and a glossary defining the terminology commonly used in VLA and generalist robotic policy research. The findings reveal substantial diversity among VLA models in terms of their supported modalities, robotic embodiments, training strategies, and architectural designs. Despite the rapid growth of VLA research, several important areas remain underexplored, including the execution of complex, long-horizon tasks, effective integration of speech, and deployment on low-cost hardware, while ensuring robust, safe, and secure operation. Full article

To address the challenge of rapid and precise obstacle avoidance for unmanned aerial vehicles (UAVs) in complex urban environments, rugged canyons, and other unstructured environments, this paper proposes a vision-based navigation algorithm. By combining the strengths of deep reinforcement learning (DRL) and convolutional neural networks (CNNs), this algorithm enables efficient navigation and obstacle avoidance in dynamic environments. First, to improve training efficiency, an autoencoder is used to extract latent spatial vectors from depth images, which are then used as input features for DRL. Second, an artificial potential field (APF) is introduced into the reward function to enhance obstacle avoidance performance in dynamic environments. Third, a CNN-based adaptive mode-switching mechanism is designed to meet navigation requirements under different environmental conditions. This mechanism can automatically identify environmental features based on real-time input data and dynamically adjust the UAV’s navigation strategy. To evaluate the proposed method, simulation experiments were conducted in static and dynamic scenarios, together with a preliminary indoor flight test. Under the evaluated conditions, the proposed method achieved favorable navigation success rates and path efficiency compared with the selected visual DRL baselines. The results also indicate cross-scenario transferability to the tested environments without environment-specific retraining. Full article

Upper-limb motor dysfunction resulting from neurological disorders severely limits patients’ activities of daily living and social participation. Pneumatic upper-limb rehabilitation robots have emerged as a promising intervention owing to their inherent compliance, lightweight design, and high power-to-weight ratio, which facilitate safe, repetitive, and home-based training. Despite these advantages, extensive clinical translation remains hindered by challenges including actuator hysteresis, nonlinear dynamics, limited accuracy in intention recognition, and inconsistent clinical evaluation metrics. This review systematically examines recent advancements in pneumatic upper-limb rehabilitation robots across four critical dimensions: structural design, human–robot interaction, control strategies, and clinical translation. We comparatively analyze rigid exoskeletons, soft wearable devices, and rigid–soft hybrid configurations based on output capability, motion accuracy, comfort, and clinical applicability. The findings suggest that while rigid systems offer high precision and soft systems maximize safety, rigid–soft hybrid architectures represent a critical developmental trend for balancing motion accuracy with interaction compliance. Furthermore, the review evaluates multimodal sensing techniques (e.g., EMG, EEG, and IMUs) for motion intention decoding and training state monitoring, alongside conventional, adaptive, and artificial intelligence-driven control methods aimed at compensating for pneumatic nonlinearity and improving real-time response. Current clinical evidence indicates that these systems effectively enhance upper-limb function and muscle strength, particularly in post-stroke rehabilitation; however, existing trials are frequently constrained by small sample sizes, short interventions, and heterogeneous protocols. Future research must prioritize rigid–soft hybrid architectures, robust multimodal sensor fusion, digital twin-assisted assessment, adaptive intelligent control, and standardized home-based rehabilitation platforms. Ultimately, this comprehensive review provides a concise reference for the design optimization and clinical deployment of next-generation pneumatic rehabilitation systems. Full article

This paper proposes a hierarchical control architecture that is established at both lower limb joint and platform levels based on a simplified admittance model, where compliance is regulated through virtual mass and damping. At the lower limb joint level, admittance control governs lower limb motion tracking, while at the platform level it adjusts the omnidirectional mobile platform velocity in response to human interaction forces. Within this framework, a Sarsa-based reinforcement learning agent dynamically optimizes the parameters of a Sigmoid function using dual state inputs. Based on hip joint angle error and human–robot interaction force, the controller dynamically adjusts virtual mass and damping to optimize the trade-off between tracking error and dynamic compliance. The simulation and experimental results on the prototype system demonstrate that, compared with traditional Sigmoid parameter-tuned admittance control, the proposed approach significantly enhances gait smoothness (dimensionless squared jerk reduced by 65.62% and 36.74% for hip and knee joints), and increases human–robot interaction compliance (RMS interaction force was reduced from 3.2502 N to 2.5109 N; EPUD decreased from 12.14 to 9.53). Moreover, the proposed strategy achieves smaller maximum overshoot (0.45° vs. 0.9°) and faster settling time (2.6 s vs. 4.59 s). These findings indicate that integrating reinforcement learning with Sigmoid parameter adaptation provides a systematic and effective solution for adaptive compliance regulation in mobile exoskeleton systems, enhancing adaptability, safety, and functional relevance for stroke patients undergoing lower limb rehabilitation training. Full article