This paper investigates the stabilization problem of a class of Cohen Grossberg BAM neural network with time-varying delays. Instead of using matrix measure methods, linear matrix inequality (LMI) methods, and integral inequality methods, a new research approach is adopted: the differential inequality method, which utilizes the properties of higher-order polynomials. By finding two local extrema points of a function and designing an appropriate controller, based on Lyapunov function theory, the asymptotic stability criterion for CG-BAMNNs is obtained. In the application of the differential inequality method, the process of finding two local extrema points and the application of the properties of higher-order polynomials are very proficient, resulting in sufficiently novel outcomes. Finally, the reasonableness and effectiveness of this approach are verified through simulation examples.
In this paper, we investigate the anomaly detection problem within stochastic networked cyber-physical systems. We introduce an anomaly detection scheme centered on estimator design employing a statistical approach. Specifically, we extend the statistical method recently developed for single systems to networked control scenarios. Additionally, we analyse the detection of several representative anomalies through this statistical approach. We then apply our anomaly detection strategy to a power network system exposed to such anomalies. Our findings demonstrate that the proposed design effectively detects these anomalies, which simpler methods relying solely on residual assessment with a threshold fail to identify.
The most crucial feature of cooperative robot is interaction safety. Ensuring safe interaction with humans is essential for improving the quality and efficiency of industrial robot production. The promotion and development of cooperative robot heavily rely on ensuring safe human-robot interaction. Therefore, this paper introduces a vision-centric cooperative robot interaction system designed for dynamic environments. Firstly, the system utilizes a highly accurate 3D camera to reconstruct the monitoring scene in three dimensions. Simultaneously, the camera mounted on the robot's head reconstructs objects within its current field of view with high precision. By merging the point clouds of these two scenes, the system achieves a comprehensive reconstruction. Secondly, the system extracts the point cloud data of human bodies in the scene and determines a safe distance between humans and the robot based on the human point cloud. This analysis allows the system to define an appropriate safety range. Finally, the system evaluates whether the human body is within the predefined safe distance from the robot and facilitates improved human-robot interactions accordingly. By implementing this approach, the system effectively ensures interaction safety in the work environment. Experimental results demonstrate the effectiveness of the proposed system in maintaining a secure interaction environment.
The integration of Generative Artificial Intelli-gence (GAI) into Digital Twins (DTs) marks a revolutionary stride in the evolution of virtual replicas for physical systems. This paper explores the cutting-edge advancements brought about by the incorporation of GAI technologies, specifically Large Language Models (LLMs), into DTs. These technologies herald a significant transformation, propelling DTs beyond their current capabilities to become more dynamic, predictive, and interactive tools that can simulate complex scenarios and anticipate future conditions with remarkable accuracy. By systematically examining the levels of GAI integration within DTs, this study delves into the methodologies and strategies for embedding AI capabilities into these virtual models. It outlines how GAI can enhance the functionality of DTs, enabling them to generate synthetic datasets, simulate unprecedented events, and provide actionable insights with LLM-based agents for decision-making. Furthermore, the paper highlights the extended applications of DTs, enriched by GAI, across various domains such as healthcare, urban planning, and beyond. The implications of this integration for operational efficiency, innovation, and decision-making processes are profound. By offering a comprehensive overview of the current state of technology and projecting future trends, this paper aims to provide stakeholders with a deep understanding of the syner-gistic potential between GAI and DTs. It sets the stage for a new era of DT technologies, where the boundaries of what can be achieved with virtual models are continually expanding.
Networked multi-agent reinforcement learning (NMARL) is widely used in multi-agent systems (MASs). However, most existing NMARL algorithms share the global state and reward, which hinders their scalability in large-scale MASs. To make NMARL applicable to large-scale MASs, we propose a proximal policy optimization (PPO) based fully decentralized NMARL. First, we design a fully decentralized multi-agent reinforcement learning (MARL) framework, formulated as a networked partially observable multi-agent markov decision process (N-POMDP). The networked MAS is represented by a graph, where each agent communicates with and shares reward only with neighbors. Second, we design a gate recurrent unit (GRU) based communication strategy to learn the temporal communication correlation. Each agent exchanges observation information and hidden state with its neighbors. Finally, we conduct experiments using the multi-agent particle environment (MPE) and compare our algorithm with common MARL algorithms. The experimental outcomes reveal the exceptional performance of our algorithm in terms of both cumulative return and convergence speed within the extensive MAS.
Currently, there is much literature discussing the smart campus framework, including terms, definitions, and practices used in campus. However, there are still many campuses facing issues with poor user experience, absence of integration platforms, inadequate data governance, etc. These obstacles have a direct impact on the overall quality and operational efficiency. Hence, a review is necessary to explore the suitable framework for smart campus. This study reviews the technology development of the smart campus and a human-centered framework is proposed to incorporate physical, cyber, and social parts to identify the essential elements of a smart campus. The proposed framework's feasibility and effectiveness are demonstrated by a real-world application. Overall, this paper contributes to the discussion on the current development of smart campuses and offers practical insights for their construction.
This paper investigates a distributed multi-target tracking (DMTT) method based on random finite sets (RFS) in distributed wireless sensor networks (DWSNs) with packet dropout. Based on the local cardinality probability hypothesis density (CPHD) filter, a consensus fusion method with information compensation (CF-IC) is proposed. The proposed method compensates the lost target information based on prior information and motion model, when some targets are lost. Finally, the performance of proposed CF-IC is evaluated through simulation experiments.
In response to the extended usage of tactical- and theater-level unmanned aerial systems (U ASs) for reconnaissance and surveillance in the modern battlefield, the ground forces put more and more emphasize on hiding their assets using camouflage or exploiting terrain and vegetation so that these assets are not visible using regular day/ $n$ ight vision sensors. The aerial multispectral (MS) imaging technology is considered as a promising remedy to rectify this situation. This paper explores the performance of commercial-off-the-shelf MS imaging sensor integrated with a small UAS to collect imagery in multiple spectral bands to potentially contribute to the detection of camouflaged targets and battlefield anomalies. In doing so it introduces a new spectral index and examines its effectiveness compared to several other methods.
Long-term operation of the power converter causes flexible changes in its state, and the model predictive control (MPC) always fails to update the predictive control model and weights to guarantee control performance. Considering the conventional digital twin (DT) structure, the virtual model is hard to simulate the physical entity with high fidelity, establishing a highly matchable predictive control model. This paper proposes a model predictive control method of Superbuck converter based on digital triplet through the optimized weight factors. The proposed digital triplet constructs a parallel triplet between digital devices in the virtual twin and physical entity to enhance the fidelity of the overall DT system. It constructs the predictive control model and weights based on the dual optimization of the virtual twin and parallel triplet, achieving a higher fidelity for the physical entity. Through the comparison of control performances on Superbuck converter between conventional MPC and DT, the proposed digital triplet empowers a more reliable and superior controller.
Efficient task planning is pivotal for multi-UAV systems navigating dynamic environments. Traditional task planning methods face challenges in adapting to the constantly changing scenarios. The emergence of large language models (LLMs) offers promising solutions to bridge this gap. Our proposal, TPML, leverages LLMs as a command interface to comprehend operators' intentions and translate them into executable codes. Harnessing the creative capabilities of generative models, TPML can command multiple UAVs in both synchronous and asynchronous patterns with a single natural-language input. Experimental results are initially validated in a tailored simulation environment before transitioning to practical implementations. Successful demonstrations of both synchronous and asynchronous missions in real-world scenarios underscore the efficacy of TPML.
This paper addresses the problem of distributed Nash equilibrium seeking in N-player games for single inte-grator dynamics subject to strongly connected networks and communication delays. First, we propose a distributed estimator for each player, enabling them to estimate the actions of all players. Notably, we take into account unknown bounded time delays that occur during communication between players and their neighbors. Next, we design a distributed Nash equilibrium seeking law using the gradient play technique. Then, we analyze the stability of the closed-loop system, which consists of an interconnected nonlinear subsystem and a linear time-delay subsystem. By means of designing the Lyapunov-Krasovskii functional, we demonstrate that Nash equilibrium seeking is achieved at an exponential rate, even in the presence of unknown and bounded communication delays. Finally, we provide a simulation example to illustrate the effectiveness of our proposed approach.
This paper introduces a synchronous generator modeling method based on neural controlled differential equations (neural CDEs) using online sampled data. This method begins with a fifth-order generator model, where every trainable parameter is regarded as one parameter of the fifth-order generator model. The objective is to use the real-time data to learn these parameters. A training algorithm has been formulated, and it has been shown that the combination of the proposed neural CDEs and the fifth-order model can produce desired online parameter estimations for the synchronous generator. The simulation results show that the proposed method can generate a very accurate estimation and model predictions with the mean absolute percentage error of 0.04638%.
In the control task of high precision tracking system (HPTS), the line-of-sight (LOS) stabilization is severely affected by disturbances and measurement noises. Inspired by the successful use of active disturbance rejection control (ADRC) in HPTS, based on the framework of cascade control, we propose a fuzzy system-based ADRC strategy. Besides, by rationally designing the inner-loop velocity controller we realize the equivalent reduce-order of the position-controlled plant and the ADRC-based strategy is taken in the outer loop to improve the disturbance rejection ability. Compared with ADRC, the proposed method takes the structure of the cascade control which has excellent disturbance rejection ability by stacking the disturbance rejection ability of inner and outer loops. Considering the measurement noises, by introducing a fuzzy system in the ADRC, the outer loop controller is transformed into a novel fuzzy self-tuning structure to filter high-frequency noise. Several experiments and simulations
Multi-unmanned agent collaboration technology is a key aspect of operations. How to reasonably assign tasks to unmanned agents before operations to maximize the overall benefits is a long-standing problem for researchers. This study systematically addresses the above issues and establishes a multi-agent task reconstruction model based on the reconnaissance, strike and assessment task background. This model includes two types of task allocation algorithms. One is task pre-allocation, which is a static task allocation before task execution. A Multi-Gene Genetic Algorithm (MGGA)is proposed for this purpose. The other is Sequential Auction Algorithm(SAA), which deals with the reallocation of tasks in response to unexpected situations during task execution. Experimental results show that, compared with other algorithms, MGGA can get the lowest fitness value under various experimental conditions. Meanwhile, SAA can handle task reallocation problems caused by various dynamic events.
Traditional visual Simultaneous Localization and Mapping (SLAM) techniques are difficult to obtain effective information in non-ideal environments such as changing light or full of smoke, which leads to the performance degradation of SLAM algorithms. To overcome the aforementioned challenges, this paper proposes a visual SLAM front-end system based on infrared-visible light fusion. The system achieves precise optimization of camera poses and map point locations in non-ideal environments by jointly optimizing the reprojection errors of visible light image point features and infrared image edge features. In addition, this article further improves the robustness of the algorithm in non-ideal environments through back-end optimization of infrared-visible light and Inertial Measurement Unit (IMU) tight coupling.
During sea-based wind turbine installation and maintenance, the ship's movement is affected by wind and waves, especially the installation accuracy and safety of the suspended wind turbine is disturbed by the non-stationary ship's roll motion. Stabilizing ship lifting platforms in complex sea conditions, a VMD-GRU-EC based prediction model for non-stationary ship roll motion is proposed. The ship movement is predicted by this model first, and then wave compensation devices are supported in the experiments to ensure safe and accurate offshore operations. To be specific, Variational Mode Decomposition (VMD) decomposes non-stationary motion into multi-modal stationary sequences, that are then input into a multidimensional Gated Recurrent Unit (GRU) for prediction. Error Correction (EC) incorporates prediction errors to further enhance accuracy. The model is applied to simulate and predict the motion of a large engineering ship under the long peak wave spectrum and Dalian sea, and the validity of the model is verified in the hardware platform. Compared to Long and Short-Term Memory (LSTM) and AutoRegressive Integrated Moving Average model (ARIMA), VMD-GRU-EC exhibits higher prediction accuracy and generalization in level 3 similar to 6 sea states and Dalian.
This paper discusses the robust predefined output containment (RPOC) control problem for heterogeneous Euler-Lagrange systems having multiple uncertain nonidentical lead-ers. In order to solve this problem, a new kind of distributed observer-based robust predefined output containment control framework is presented. Firstly, for obtaining the information of nonidentical leaders' dynamics including uncertain parameters in leaders' system matrices and states, two kinds of adaptive observes are constructed in a fully distributed form without any knowledge of the system matrices of nonidentical leaders, exactly. Secondly, on the basis of adaptive learning technique, a new RPOC controller is then developed by using the presented observers. Furthermore, with help of Lyapunov stability theory, the RPOC criteria for the considered system under multiple uncertain nonidentical leaders are derived from the constructed controller. At last, a simulation example is provided to demonstrate the effectiveness of the proposed RPOC controller.
Air-ground Collaborative SLAM (CoSLAM) is particularly suitable for route planning and mapping in large scenes. However, accurate localization of an Unmanned Ground Vehicle (UGV) in CoSLAM has been a challenging task, making route planning for various applications not always reliable. In this study, we design a air-ground collaborative system, which applies Ultra Wide Band (UWB) ranging units on both Unmanned Aerial Vehicle (UAV) and UGV as additional sensors to collect range information between the UAV and the UGV. The range information is used to correct the position of the UGV provided by the UAV. The correction is realized through an extended Kalman filter. Simulation results verify the proposed method.
Due to the small size of 3C (Computer, Communication and Consumer electronics) parts and their high assembly accuracy requirements, traditional industrial robots cannot meet the needs of flexible assembly. To address manufacturing errors and uncertainties in part poses, this study proposes a markerless visual servoing method. Through analysis of assembled 3C parts and manual assembly movements, a high precision and flexible assembly strategy has been designed. This strategy divides assembly into into long-range approaching phase and a short-range aligning phase. To adapt to these phases, a quotient kinematic manipulator (QKM) consisting of two parallel manipulator modules was developed, along with a well-designed camera hardware system. In the long-range approaching phase, an image moment-based visual servoing method with constrained model predictive control is employed, addressing robot motion collisions and end-effector velocity constraints, resulting in collision avoidance motion trajectories. In the short-range aligning phase, a weighted image brightness visual servoing method is proposed, utilizing surface texture features to replace geometric features, achieving high precision alignment of components. Experimental results show the proposed servoing assembly strategy achieves high-precision collaborative manipulator assembly without artificial markers, with accuracy of 0.5mm/0.5 degrees.
An Unmanned Aerial Vehicle (UAV) lifeguard scout quadcopter is implemented to detect rip currents and threats. Requirements for scout UAVs are discussed and applied to the scout UAV configuration. The scout UAV is equipped with an AI-capable mission controller, triple communication and control radio links, and 6.7x9.8N thrust from actuator and propeller combination to handle additional payload such as large-capacity battery for extended flight duration. The scout UAV conducted multiple flight missions to collect 11.8k ocean images with 10.9km flight mileage. Using the collected images, optical flow is analyzed, and a new information-weighted scaling with image gradient density is developed and demonstrated for a reliable optical flow result. A channel current flow analysis with depth and risk models is proposed to estimate net channel current risk and to detect rip currents.
