The evolution of Artificial Intelligence (AI) has reached a critical point, where agentic AI systems demonstrate strong capabilities in goal formulation and planning but remain difficult to deploy in real-world settings due to their limited grounding in physical environments. These limitations arise from the challenges of partial observability, actuation uncertainty, and strict resource constraints that characterize the physical world. This survey argues that the Artificial Intelligence of Things (AIoT) provides the necessary foundation to embed agentic intelligence into such environments by enabling continuous interaction between sensing, reasoning, and action. We analyze the synergy between goal-driven agentic AI and distributed AIoT infrastructures and present a unified taxonomy of AIoT-enabled agentic architectures, highlighting trade-offs across centralized, edge-native, and hybrid deployment models. The survey further examines key enabling technologies, including edge intelligence, semantic communication, digital twins, and trust mechanisms, and discusses how they integrate into cognitive control loops. Through representative applications in smart cities, industrial automation, healthcare, and energy systems, we show how this convergence moves automation beyond rule-based behavior toward context-aware autonomy. Finally, we identify open challenges related to long-horizon safety, resource-aware intelligence, and ethical governance, and outline research directions toward robust, trustworthy, and socially embedded autonomous systems.

Detection of lung nodules is key in the treatment of early-stage lung cancer. Computed tomography (CT) scanning technology is an essential contactless tool. However, stray radiation caused by a patient's slight movements and equipment operation can impair CT images, hindering accurate lung nodule detection. To address these issues, this study proposes an artificial intelligence-based anti-interference lung nodule detection method, which is primarily structured with Yolov8 and combines the modules of adaptive gating sparse attention (AGSA) and haar wavelet downsampling (HWD), referred to as Yolov8-AH. This model aimed to improve the accuracy of lung nodule detection in lung CT images under interference conditions. AGSA focuses on key areas of the image, promoting detection stability even when CT images are disturbed. Furthermore, HWD prioritizes the frequency components corresponding to the size and shape of the nodules, enhancing their visibility for easier detection and analysis. HWD effectively reduces image noise without significantly blurring the lung nodule edges, emphasizing them prominently within the lung tissue. Furthermore, when combined with the Yolov8 deep learning model driven by artificial intelligence, the model could accurately detect lung nodules, significantly aiding in early diagnosis and treatment. The effectiveness of the Yolov8-AH detection model was verified through ablation experiments, experiments under varying noise intensities, and experiments under different noise application ratios. The experimental results demonstrate that, compared to existing lung nodule detection models, the Yolov8-AH model achieves a 24% improvement in mAP50 and an 8.2% improvement in precision.

As sustainable development gains attention, integrating carbon-intelligent computing into fault diagnosis systems has emerged as a critical strategy to reduce energy consumption and carbon footprints. This approach uses artificial intelligence (AI) and the Internet of Things (IoT) to optimize task scheduling, aligning it with low-carbon energy sources based on time and location. In fault diagnosis, energy-intensive tasks, such as data processing and model inference, can be scheduled during periods of abundant renewable energy, thereby minimizing emissions. However, drilling machines operate under complex conditions that generate nonstationary noise, which distorts signals and complicates fault diagnosis. Therefore, this article combines bidirectional long short-term memory (BiLSTM) with the Kolmogorov-Arnold network (KAN) and integrates Wavelet Transform and Convolutional Autoencoder, proposing a highly robust fault diagnosis model for drilling machines, named WCBK. The Wavelet Transform converts pressure time-series data, which contains fault information, into time-frequency images, facilitating the detection of fault frequency components. The Convolutional Autoencoder preserves essential features while removing noise by learning low-dimensional representations of the signal, effectively capturing local features in time-frequency images through local connections to enhance denoising performance. Finally, the composite deep learning network, which combines BiLSTM and KAN, achieves highly robust fault diagnosis under complex working conditions. The effectiveness of the proposed WCBK model was validated through ablation experiments, experiments on different individuals, experiments on different parts, and model adaptability evaluations. In experiments involving different individuals and parts, the WCBK model improved fault diagnosis accuracy by 10.9% and 8.8%, respectively, compared to existing models.

Traditional traffic anomaly-based attack detection methods in Zero-trust Networks (ZTN) suffer from inherent security vulnerabilities, as they neglect considerations regarding their security defenses. Compromising the attack detection model itself can result in the breakdown of normal attack detection capabilities. Ensuring the security of the attack detection model during runtime presents a novel challenge. To address these shortcomings, we propose a novel attack detection model, termed Security within Security: Attack Detection Model with Defenses Against Attacks Capability for Zero-Trust Networks (SWS), aimed at enhancing the security of ZTN. SWS focuses on achieving attack detection in non-secure detection environments, to maintain its detection capability even when under attack. By employing a soft thresholding method, SWS adapts to the dynamic changes in network traffic, thus reducing the interference of attack signals. The incorporation of an attention mechanism enables SWS to concentrate on analyzing the most indicative traffic features of attack behavior. Additionally, we integrate Residual Networks (ResNet) and Bidirectional Long Short-Term Memory (BiLSTM) to enhance the robustness of identifying complex network attack behaviors. The effectiveness of the SWS is validated through ablation studies, model comparisons, experiments conducted over different training epochs, and experiments conducted on various components of the dataset. Experimental results demonstrate that compared to existing attack detection models, SWS achieves improvements in detection accuracy and recall rate by 13.4% and 10.6%, respectively, while reducing the False Positive Rate (FPR) by 16.9%.

Predicting vehicle trajectories is a crucial component of intelligent transportation systems, bearing significant research significance. Leveraging the latest advancements in deep learning and data processing technologies enables us to model intricate interactions among multiple vehicles in complex traffic scenarios. Traditional trajectory prediction methods typically rely on sensor data and vehicle behavior models, which may struggle to capture the intricate relationships between vehicles in dynamic, high-traffic situations, as well as the topological complexities of road networks. To tackle these challenges, we introduce a novel approach: the Vehicle-Driven Dynamic Graph Neural Network (V-DGNN) model. This model starts by constructing an interaction graph among vehicles, enabling the simultaneous capture of both temporal and spatial dependencies between them. Moreover, it incorporates a spatiotemporal attention network for extracting vehicle motion patterns. We also propose a unique mechanism to address the challenges posed by high-speed spatiotemporal changes. This mechanism involves the sensitive sampling of nearby timestamps to effectively learn the dynamic distribution of vehicles. Additionally, the model incorporates vehicle behavior features and road network topology information as supplementary inputs while minimizing prediction variances. This equips our model with the ability to make robust predictions even in the face of distribution changes. Experimental results on two real-world datasets convincingly demonstrate that our approach outperforms current state-of-the-art models, delivering superior long-term predictive performance.

With the rapid development of Internet of Vehicles technology, mobile communications are gradually integrated with various fields, and more and more Internet of Vehicles equipment are connected to the Internet. However, existing online information transmission methods mainly rely on the original network infrastructure. Once the communication infrastructure fails, it is likely that information transmission will fail or even be lost. In this paper, we utilize the optical modules that come with sensor nodes to implement a hybrid communication debugging system based on Visible Light Communication (VLC). To enhance uplink reliability, we’ve devised an optical camera-compatible frame synchronization method. Leveraging the Transformer algorithm, we predict frame header positions, thereby bolstering data collection reliability. Additionally, for efficient debugging information uploading, we logically group and organize the initial data, and use the Snappy compression algorithm to decrease empty time slot count to complete the data compression, saving time. Finally, confidentiality enhancement technology is introduced in the physical layer, and a new security enhancement optimization scheme based on Artificial Noise is proposed. The Artificial Noise (AN) sent by the sender enables the sender to counter eavesdropping interference, and the authorized recipient can cancel the Artificial Noise (AN). The results show that the proposed scheme is more secure.

UAV technology has been developing rapidly in recent years, and UAV remote sensing image target detection plays an important role in military, agriculture, forestry and marine fields. Currently there is a lack of consideration for the safety aspects of UAV image target detection, so we propose an intelligent interference information fusion method to expand the dataset. By adding interference to the original dataset images and generating the affected dataset images to train the model, the anti-interference ability and robustness of the target detection model are improved. We also propose a highly robust target detection framework (RA-RTDETR) with strong anti-interference capability to improve the detection accuracy of disturbed images. The framework fuses three methods, deep residual contraction network and cascading attention mechanism, and is experimentally compared with other methods (yolov5s, yolov8s, RTDETR) on a dataset. The experimental results demonstrate that the dataset with fused interference information has a significant improvement in model robustness, and the accuracy of our proposed RA-RTDETR achieves the best results.

Given the severity of air pollution, air quality monitoring has become a crucial aspect of Artificial Intelligence of Things (AIoT) applications, providing essential information for forecasting air pollution. However, the training process for air quality monitoring models heavily relies on the high-performance computing resources, leading to significant energy consumption and associated carbon emissions. This contradicts the objectives of low-carbon and sustainable computing. This article proposes a new hybrid PM2.5 prediction model (NHPPM) for air quality monitoring to address the above challenges. NHPPM prioritizes energy efficiency while maintaining high prediction accuracy by integrating several power-efficient strategies. First, Wiener filtering is used to denoise the multisource air quality data enhancing the efficiency of the multisource data fusion. Second, variational mode decomposition (VMD) decomposes different components of the multisource air quality data, helping to identify and separate the most important factors affecting pollutants. This reduces the data needed for model training and leads to lower resource consumption. Kernel principal component analysis (KPCA) transforms the high-dimensional data into a lower-dimensional representation while retaining the critical information, further minimizing computational demands. Additionally, this article utilizes the informer deep learning model to analyse the trends in air quality data. The model’s effectiveness is validated through the ablation studies, performance evaluation experiments, and short- and long-term prediction experiments. The experimental results show that our model reduces the mean absolute error (MAE) and root mean-square error (RMSE) by 16.2% and 14.9%, respectively, compared to the existing PM2.5 prediction models. Furthermore, it reduces the energy consumption of the model training by 33.8%.

In the context of the booming Internet of Vehicles, predicting vehicle trajectories is crucial for intelligent transportation systems. Existing methods, reliant on sensor data and behavior models, struggle with intricate relationships between vehicles and dynamic road networks. To overcome these challenges, we propose the Vehicle Interaction-based Dynamic Graph Neural Network (VI-DGNN) model. This model constructs a vehicle interaction graph to capture temporal and spatial dependencies among vehicles. A spatiotemporal attention network is employed to discern patterns in vehicle movements, addressing high-speed changes. Our model introduces a vehicle interaction mechanism for dynamic movement, leveraging proximity timestamp graph structures. By incorporating vehicle behavioral features and road network topology, our model minimizes distribution prediction variance, enhancing stability. Experimental results on real datasets demonstrate superior long-term prediction performance compared to state-of-the-art baselines.