In recent years, several countries, such as China and the United States, and high-tech companies, such as Google, have increased investment in artificial intelligence. Deep learning is one of the current artificial intelligence research key areas. We analyze and summarize the latest progress and future research directions of deep learning. First, we outline three basic models of deep learning, which are multilayer perceptrons, convolutional neural networks, and recurrent neural networks. On this basis, we further analyze the emerging new models of convolution neural networks and recurrent neural networks. Furthermore, we summarize the applications of deep learning in many areas of artificial intelligence, including speech processing, computer vision, and natural language processing. Finally, we discuss the existing problems of deep learning and provide the corresponding possible solutions.

This work studies the overall control problem of a multiterrain adaptive wheel-legged inverted pendulum robot. The dynamic model of the robot is also established. The linear quadratic regulator (LQR) algorithm analyzes the decoupled balance and longitudinal motion subsystems, and the controller is designed. Based on virtual model control, the torque in the inverted pendulum robot is converted into the joint torque in the wheel-leg structure. The controller's performance is simulated by building a simulation platform (Simulink Mulitibody). The corresponding controller is designed to control the robot's height and roll attitude. The performance of the whole controller is verified in the robot, which has certain theoretical and practical values.

The study of swarm intelligence, inspired by self-organized behaviors in biology, has achieved remarkable momentum. By understanding the intrinsic mechanisms of self-organization behavior in biological systems, researchers can design efficient swarm robotics systems. The research achievements in swarm robotic systems based on the self-organization behavior in biology are comprehensively assessed in this study. First, several self-organization behaviors in biology are discussed. Second, the studies on the tasks of swarm robotic systems derived from behavioral mechanisms are categorized, and the achievements in recent years are summarized. Finally, the key challenges and the future development direction in the growth of swarm robotic systems inspired by biological behavior are explored. This study aims to offer researchers a novel perspective for contemplating the research outcomes of collective intelligence swarm robotic systems, thereby promoting advancements in related fields.

The generation of safety-critical scenarios is a pivotal focus in the domain of autonomous driving, holding significant application value in areas such as autonomous driving testing, automotive safety assessments, and the establishment of automotive safety standards. It is the key to the implementation of autonomous driving applications. Existing research lacks a survey focusing on safety-critical scenario generation techniques. We provide a systematic review of safety-critical scenario generation techniques. We summarize the research progress in the field of safety-critical scenario generation techniques. Furthermore we conduct a comparative analysis of models dedicated to safety-critical scenario generation. In addition we explore safety-critical scenario generation methods based on clustering Bayesian networks and adversarial networks. Finally we present a prospective outlook on research trends in safety-critical scenario generation methods.

The cable-driven flexible manipulator has several advantages, such as high deformation, small mass, and low energy consumption. With the development of robotic applications, the precise modeling and effective control of cable-driven flexible manipulators have become a key research direction for domestic and foreign scholars. This study reviews cable-driven flexible manipulators′ modeling and control methods; analyzes their kinematic, dynamic, and static modeling from the modeling method; and summarizes the control methods from model-based and model-free perspectives. The model-based control methods are divided into kinematic and dynamic modeling. Meanwhile, the model-free control methods include fuzzy, neural network, adaptive, and sliding mode control. Finally, the future research direction of modeling and motion control of cable-driven flexible manipulators prospects.

The simultaneous localization and mapping (SLAM) algorithm is the key link in achieving the autonomous mobility of mobile robots. Light detection and ranging (LiDAR) has the advantages of high range accuracy, less susceptibility to external interference, and intuitive and convenient map construction, and is widely used in map construction for large and complex indoor and outdoor scenes. Domestic and foreign scholars have achieved fruitful results in the research of SLAM algorithms based on 3D LiDAR due to the application and popularity of 3D lasers. The current status of domestic and foreign research on 3D laser SLAM algorithms in front-end data association, back-end optimization, etc., and the principles and advantages and disadvantages of various 3D laser SLAM algorithms and improvement schemes are analyzed and summarized in combination with deep learning and multi-sensor fusion. The application of theories and technologies in 3D laser SLAM algorithms is described, and research hotspots and development trends of 3D laser SLAM algorithms are highlighted, including multi-source information fusion, integration with deep learning, the robustness of application scenarios, a generic framework for SLAM algorithms, and technology penetration of mobile sensors and wireless signal regimes. The research results have significant reference value and guiding significance for the research of 3D laser SLAM algorithms and instant localization and map construction of mobile robots in unknown environments.

In recent years, both industry and academia have made significant advances in deep learning (DL). However, configuring the hyperparameters of deep models typically requires significant computational overhead and expert knowledge. To overcome these aforementioned challenges, evolutionary computation (EC), as an efficient heuristic search, has demonstrated significant advantages in the automated configuration of DL models, i.e., evolutionary DL (EDL). We describe EDL from the perspective of automated machine learning. Particularly, we first depict the concept of EDL from EC and DL perspectives and regard EDL as an optimization problem. Consequently, we systematically introduce data preparation, model generation, and model deployment from the DL lifecycle. In addition, we analyze and discuss the solution representation and search paradigms. Finally, we provide applications, open issues, and potential research directions related to EDL. This study reviews the advancements in EDL and offers insightful guidelines for its development.

With the emergence of complex economic and social decision-making problems and the increasing complexity and uncertainty of the decision-making environment in China, traditional group decision-making theories and methods are undergoing profound change. Large-group decision-making theory and methods have developed and become hot fields in quantitative decision-making research with broad application prospects in practical decision-making scenarios. In this paper, we review current research and development in large-group decision-making from three main aspects: Multi-attribute, conflict-related, and risk-related large-group decision-making methods. We then introduce research hotspots and the latest progress in the field of large-group decision-making, and discuss its problems and challenges. Lastly, we note the future development trends in large-group decision-making.

In recent years, due to unmanned aerial vehicles (UAV) characteristics such as flexibility, intelligence, and autonomy, UAV has been widely used in military and civilian applications, especially the primary target tracking task in search and surveillance. However, due to the complexity of the scene environment and the variability of the moving targets in UAV visual target tracking, it is difficult to extract features and formulate a model for the target, bringing a significant challenge to the tracking performance. Therefore, we provide a review of UAV visual target tracking. First, we present the current state of researching visual target tracking, review the classical and latest tracking algorithms, especially those based on correlation filtering and deep learning, and compare the advantages and disadvantages of different algorithms. Then, we summarize the commonly-used UAV target tracking datasets and evaluation metrics. Finally, we present a prospect of the future directions for developing UAV visual target tracking algorithms.

Flexible manipulators usually have flexible joints ard/or links. Over the last four decades, research on flexible manipulators has greatly progressed. In this paper, the development status, research hotspots, and frontier progress of flexible manipulators in dynamic modeling methods, vibration sensor measurement systems, and control algorithms are reviewed. Firstly, the advantages, disadvantages, and complexities of flexible manipulators are briefly explained, and the modeling methods and technologies of flexible joints and flexible manipulator robots are introduced. Secondly, the application and characteristics of different vibration measurement systems in flexible manipulators are described. The advantages and disadvantages of different measurement methods in the vibration measurement of flexible manipulators are compared. Then, the vibration control methods and control algorithms of flexible manipulators proposed in various studies are emphatically reviewed. The research status of the self-excited vibration and control of flexible parallel manipulators is also analyzed. Lastly, the development trend, problems, and challenges of future research work are discussed.

With the development of industrial Internet technology, the data volume of industrial equipment has increased exponentially, which has brought a huge data pressure to cloud computing. There is more and more applications and researches of edge computing in the industrial field due to the characteristics of low latency, low traffic, and good privacy. We review the research on cloud computing and edge computing in industrial applications in recent years. We firstly introduce the development history of cloud computing and edge computing in the context of the industrial Internet, and analyze the definition of edge computing and the relationships between several typical forms of edge computing and industrial edge computing system. Secondly, several typical applications of industrial edge computing are analyzed. In view of the current research status, several key technologies affecting edge computing applications in industrial scenarios are discussed. Finally, the research challenges in the industrial scene are summarized and prospected.

Multispectral object detection technology is an approach that utilizes information derived from diverse wavelength spectra to detect and recognize objects. This technology integrates data of different spectras from visible light, infrared, and thermal imaging, and improves the accuracy and robustness of object detection. This technology performs excellently in low-light or other challenging environments. This technology is commonly applied in fields such as night vision monitoring, meteorological observation, agriculture, and environmental monitoring. First, the definition, acquisition methods, and common datasets of multispectral images are explained. Second, multispectral object detection algorithms are categorized, and examples of relevant algorithms are given. Third, the applications of the algorithms, which center on specific application cases in pedestrian detection, agriculture, and environmental monitoring, are discussed. Finally, the future directions for multispectral object detection are explored.

Driven by current demand, wireless sensor network localization has become a hot research field. First, in this paper, the research on wireless sensor network localization is summarized from three aspects: localization method, localization technology, and localization principle. Second, three localization algorithms based on wireless sensor networks are described. Third, the hot issues on wireless sensor network positioning are described. Aiming at the non-line-of-sight (NLOS) interference problem in localization research, the research status of the NLOS error is analyzed from two aspects: numerical and statistical characteristics. In addition, the localization methods of wireless sensor networks based on multi-sensor fusion are analyzed and summarized. Finally, the challenges of applying wireless sensor network localization are discussed.

The event-based camera is a novel bio-inspired vision sensor that efficiently captures scene changes in real-time. Unlike traditional frame-based cameras, event cameras only report triggered pixel-level brightness changes (called events) and outputs asynchronous event streams at microsecond resolution. This type of vision sensor has gradually become a hot topic in the fields of image processing, computer vision, robot perception and state estimation, and neuromorphology. We first describe the basic principles, developmental history, advantages, and challenges of event cameras. Then, three typical event cameras (including the DVS, ATIS and DAVIS) and multiple advanced event cameras are introduced. Next, we review the research on applications of event cameras, including feature extraction, depth estimation, optical flow estimation, intensity image estimation and 3D reconstruction, object recognition and tracking, autonomous localization and pose estimation, visual odometry and SLAM, multi-sensor fusion, and other aspects. Finally, the research progress of the event cameras is summarized and the future development trend is discussed.

As an outgrowth of Industry 4.0, "zero-defect manufacturing" is dedicated to dramatically improving product yield and ultimately achieving zero-defect products. Presently, the manufacturing process mainly adopts the physical inspection method for product quality inspection, which is an off-line test with high detection cost and delayed guide. By monitoring production process data and predicting product quality or process, virtual metrology (VM) may transform the traditional off-line and delayed quality sampling into online and real-time quality full inspection. Firstly, we summarize the development of VM over time. Then the research status and typical application scenarios of VM, especially in semiconductor manufacturing, are introduced. Subsequently, the common VM techniques and practical engineering problems are outlined, such as data preprocessing, predictive modeling methods and system function design. Finally, we prospect the manufacturing process VM problem, and propose a manufacturing process intelligent management system, which integrates data preprocessing and visualization, VM and quality tracing.

A Markov jump linear system (MJLS) is a random system with multiple modes. The jump transition of the system between each mode is determined by a set of Markov chains. The MJLS model can accurately describe the system in actual engineering applications because it can produce mutations in the representation process. In recent years, the optimal control problem of the MJLS has become a research hotspot. Dynamic programming, maximum value principle, and linear matrix inequality have become the mainstream methods to solve such problems. This paper reviews the current research status of the MJLS optimal control field. The research status of MLJS optimal control problems at home and abroad under general conditions, noise conditions, time delay conditions, and some specific conditions are discussed individually. Finally, the research direction of the MJLS optimal control field worthy of attention in the future is summarized and put forward.

Multiple autonomous underwater vehicles (MAUV) are increasingly applied to marine exploration and resource development. As such, they have been widely explored. In this paper, the tasks and methods of common MAUV are discussed, and the key problem of cooperation among MAUV is comprehensively investigated from the perspective of emerging functions. Firstly, the cooperation among MAUV is classified according to task-, motion-, and measurement-space aspects and their combination. Secondly, different kinds of MAUV and their realization methods are introduced from three perspectives. Lastly, the future research direction of MAUV is discussed. This survey is conducted to provide strategies and approaches for MAUV and help accomplish underwater tasks efficiently and cooperatively in complex ocean-related application scenarios.

Intelligent communication technology has become a research hotspot with complex electromagnetic environment in electromagnetic field. It is of great significance to analyse essential radio mechanism and realize the spectrum sensing and modulation recognition tasks in complex environment. Focusing on basic and cutting-edge requirements of electronic information system under complex electromagnetic environment, we introduce research status of radio signal modulation recognition. On this foundation, we expound traditional and intelligent radio signal modulation recognition methods is with the development of electromagnetic characteristics respectively. Finally, we comprehensively summarize key technologies and look forward the development trend in this field.

We address the problem of how UAVs in the three-dimensional environment can reasonably plan a path that avoids obstacles from the starting point to a target point. UAV path planning and obstacle avoidance method are proposed, based on the combination of improved particle swarm optimization and rolling strategy methods. First, the proposed method uses the UAV as the center and uses sensors to establish a UAV visual area model.Next, scrolling strategies are combined to obtain information about the environment surrounding the UAV.Lastly, the improved particle swarm algorithm is used to search the path, with the addition of integrated corner control to improve path smoothness. Pheromone and heuristic functions are added to the traditional particle swarm optimization algorithm to enhance the algorithm's global search capability, and the parameters are specifically designed to improve the algorithm's convergence speed. Simulation results show that the proposed method can achieve real-time obstacle avoidance, the planned path is relatively smooth, and the improved algorithm has higher convergence than the traditional algorithm.

The modeling and control of quadrotor UAV-slung load system is a popular research field in recent years. Quadrotor UAV-slung load system is an underactuated nonlinear system formed by connecting the load and the Quadrotor UAV with a cable. The modeling of full state system is usually based on Euler-Lagrange equation. To make convenience for controller design, equivalent transformations and simplifications of the system model are usually adopted. According to the different simplified models, many control methods, such as PD control, feedback linearization and backstepping, are adopted, and has achieved control objectives like UAV trajectory tracking, load trajectory tracking, UAV formation planning and obstacle avoidance, etc. If there exist unmodeled dynamics or unknown parameters in system model, the controller can be improved by adaptive control method or introducing observer, etc. For the sake of enhancing robustness and flight safety, the controller can be developed based on optimum control theory, and the control performance can be further enhanced by use of input shaper and saturation function, etc. Intelligent control methods have achieved initial results in some complex control problems such as UAV path planning. Cooperative transportation by multiple UAVs is another emerging research subproject.