In this paper, a PID nested higher-order nonsingular terminal sliding mode observer is proposed for speed estimation of Permanent Magnet Synchronous Motor (PMSM). Here, the introduction of PID error embedded into the nonsingular terminal sliding manifold, which is now at higher order, smoothen the observer's control input, leading to significant chattering reduction as well as better accuracy and faster response of the resulted-in back-EMF and rotor's electrical angle estimation. The simulation and experimental results demonstrate the effectiveness and advantages of the proposed approach.
As autonomous driving gains widespread attention, extensive research is being conducted to enable robots to safely reach their destinations independently. Before the rise of advanced robotic artificial intelligence, much research focused on path planning for route exploration and planning. However, these methods have limitations in responding to dynamic environmental changes. To overcome this, efforts have been made to integrate reinforcement learning. This study proposes an adaptive path planning reinforcement learning algorithm, combining the efficiency of path planning with the adaptability of reinforcement learning to various environmental changes. The proposed method effectively integrates traditional path planning algorithms with reinforcement learning. We utilize the TD3 network as the backbone and apply the Artificial Potential Field algorithm, known for its robustness to obstacles. Experimental results demonstrate that our approach achieved approximately 14 points higher average rewards and similar maximum rewards compared to existing networks, indicating higher average success rates. Additionally, results specified above indicate faster or comparable convergence speeds relative to traditional networks.
Delta robots are parallel robots consisting of three arms connected to universal joints at the base and at the end -effector. They are known for their high -speed and precise movement, making them ideal for tasks requiring rapid pick-and-place operations, since the orientation of the end-effector is constant and usually parallel to the work surface. The design features lightweight arms and a fixed base, with intermediate joints consisting of universal joints or spherical joints, which add inertia and backlash to the system. This paper introduces a Delta robot with the use of compliant joints to minimize the number of components. The new compliant joints replicate a universal joint, reduce the inertia of the system, are easy to manufacture, and minimize backlash. The added stiffness alters the precision of the robot, but the stored potential energy in the joint increases the dynamics of the robot, which makes the robot suitable for tasks that requires speed instead of accurate positioning.
Accurate forecasting of pedestrian counts in the Central Business District of Melbourne during extreme scenarios such as the COVID-19 pandemic is crucial for optimizing resource allocation and ensuring public safety. Current models lack the incorporation of major disruptions, leading to a critical gap in urban traffic forecasting. This study addresses this gap by developing an adaptive forecasting model, namely hyperparam-eter finetuning-convolutional neural network-multivariate-long short-term memory. Using hourly pedestrian counts (2020–2022) and COVID-19 case data, historical trends are integrated with pandemic factors to capture long-term patterns and sudden disruptions. Based on three evaluation metrics (NRMSE, MAPE, and R2), the results demonstrate the accuracy of the model in predicting pedestrian traffic during disruptive events. This research advances context-aware pedestrian traffic predictive modelling, enabling informed decision-making in urban environments.
Wireless power transmission remains an open challenge for implantable devices as well as the active multifunctional capsule endoscope. This paper presents a development of high-powered wireless power transmission system for self-actuated soft capsule endoscope. The system consists of a transmitting unit and multiple receiving units. The transmitting unit is capable of generating a powerful uniform alternative magnetic field at multiple desired frequencies, while each of the receiving unit is designed to resonate with transmission signal at a desired frequency. As a result, the proposed wireless power transmission system can transfer 2.1 W power at frequency range from 70–100 kHz. This wireless power transmission system will be applied to control the morphology changing of each soft actuator to create the locomotion of the soft capsule endoscope.
In this paper, we consider the discretization of the continuous-time dynamic model and process noise that occurs in ballistic object filtering problems. Our analysis shows that existing discretization methods are ad hoc in nature. We apply the 1.5 order Ito-Taylor approximation to these discretization problems. We specifically study the ballistic projectile filtering problem and derive equations for the discretized dynamic model and process noise. Our results are compared with those from existing algorithms.
The wind turbine generator (WTG) type 4 is a promising prototype for the renewable energy industry. How-ever, its implementation is challenging due to the numerous standards, especially those for fault ride-through (FRT) capa-bilities. This paper evaluates the performance of two different FRT strategies for WTG type 4 under grid conditions, taking into account varying values of short circuit ratio (SCR) and X/R ratio. The aim is not only to gain a comprehensive understanding of these FRT strategies' characteristics but also to assess the response of WTG type 4 in grid fault. Two primary scenarios are conducted: grid voltage oscillation over a certain period and transmission line short-circuit faults. The study was validated using electromagnetic transient simulations.
The superior driving efficiency and reduced emissions of electric buses have been propelling the adoption of these vehicles in urban centers recently. However, transitioning from diesel to electric fleets presents numerous challenges, including fleet sizing and management, and charging infrastructure planning and operation. These issues complicate the decision-making process, especially given the variety of charging strategies (e.g., overnight charging, fast charging at terminal stations, and battery swap). This paper proposes a novel framework that integrates optimization and multi-criteria decision analysis (MCDA) to tackle the complexities related to planning and operating electric bus systems. The optimization model aims to minimize the capital and operational expenditures for different bus-system configurations. The results from the optimization model are then assessed using an MCDA approach, considering multiple and incommensurable criteria, to identify the best-ranked fleet configuration. In this context, this framework can aid planners to make more informed decisions in the transition to electric bus fleets.
This paper examines the finite-time convergence characteristics of higher order terminal sliding manifold for control design purposes. Here, with a set of conditions to be satisfied, a unique analytical solution for this particular type of nonlinear higher order differential equation is obtained. As a result, the convergence time is then calculated accordingly and the finite-time converging-to-zero trajectories of the states, when the sliding mode is reached, can be systematically designed.
In recent years, the deployment control of multi-agent systems has gained a lot of attention. Creating an attractive and repulsive force with an artificial potential field (APF) is one of the most often utilized techniques for arranging the agents into an equilibrium configuration. Large-scale homogenous robot deployment in planar areas is proposed in this paper using a distributed APF approach. Our framework addresses situations where the group of robots needs to spread out to maximize the coverage area and enforces the restriction that the robots form a fully connected triangular lattice network. The main idea is the deployment method inspired by the distributed Morse potential field. However, instead of directly controlling the robot velocity by employing the gradient of the potential field, the desired position based on the neighbor's localization is generated. Due to the implementation of a position controller, the method is compatible with any conventional navigation strategy when the environment is given. The framework continuously reconfigures the robot's desired position to prevent an unreliable or suboptimal network. The deployment method is evaluated by the coverage metric and mean nearest-neighbor distance metric. The compatibility with navigation systems to deploy robots in non-convex environments is also proved.
Accurate indoor target localization in Wireless Sen-sor Networks (WSNs) is challenging due to the difficulty in detecting weak direct path (DP) of channel impulse response (CIR). The current hard decision (HD) method only considers the energy of a single path, so its performance deteriorates in scenarios where weak DP is prevalent. Although the existing soft information decision (SID) method takes into account the physical structure of the DP, it relies on an accurate statistical model. To detect weak DP more accurately in practical scenarios, we propose a SID method based on scoring models (SIDSM). This method first models the signal-present and noise-only scores. The soft information (SI) of CIR is then extracted using the rule of thirds. The process of DP detection entails finding the maximum value of SI. To improve the accuracy of indoor localization, a localization method based on SI cross-correlation function (SICC) is developed. It establishes the cost function through the cross-correlation function of SI and then solves it through the exhaustive search method. The effectiveness of the methods has been verified by utilizing experimental data collected and processed via indoor WSNs. The results demonstrate that the DP detection performance of the SIDSM surpasses that of the HD method. Furthermore, the SICC method shows improved localization accuracy compared to conventional HD-based local-ization methods.
Vehicle detection from aerial images is crucial for effective traffic management and safety, especially during night-time when visibility is low and the risk of accidents is higher. However, detecting vehicles at night is challenging due to the lack of adequate training data under low-light conditions, making it difficult to develop models that are both accurate and reliable. In this paper, we address the challenge of vehicle detection in nighttime traffic environments captured by drones. Utilizing the VisDrone-DET2019 dataset, we first transform daytime images to nighttime images using CycleGAN, thereby generating a comprehensive dataset that simulates night conditions. Subsequently, we apply advanced object detection algorithms, including YOLOv8, YOLOvlO, and RTMDet, to identify vehicles within the gen-erated nighttime dataset. Our experimental results demonstrate the effectiveness and robustness of these detection models in low-light conditions, providing significant in-sights into the development of reliable traffic monitoring systems for nighttime scenarios.
The proliferation of pornographic content online challenges content moderation efforts, especially in sensitive contexts. Traditional detection methods often require extensive labeled datasets and struggle with nuanced content. This paper proposes a zero-shot classification approach using Vision-Language Models (VLMs) like CLIP and Open CLIP, which leverage visual-textual alignment to classify pornographic content without task-specific training. Evaluated on the LSPD dataset, our research examined various aspects of using VLMs, including the effects of keyword choice, prompt construction, model size, and pre-training data resolution. Our method achieved comparable or better performance than traditional models, with the best accuracy reaching 91.6% using the key-word “erotica“ and specific descriptive prompts. This approach reduces dependency on large datasets, offering a robust solution for detecting explicit content.
Advanced machine vision and deep learning models are increasingly used as virtual sensors to detect and classify targets via class probabilities. Common statistical multitarget filters and their associated information fusion methods rely on point measurements, and integrating target class measurements into those solutions is challenging. This paper introduces a statistical multitarget filter, formulated in the labeled multi-Bernoulli filtering framework to handle class measurements from virtual sensors. Applied to multi-sensor, multitarget filtering in centralized and distributed networks, numerical experiments show that incorporating class information enhances tracking accuracy, especially when shared across networks. The results highlight the benefits of using class probabilities to improve tracking in dynamic environments.
This paper demonstrates an approach to integrate object-aware map building technique which employs visual language models in microcomputers. This paper addresses the computational challenges of deploying visual language models in resource-constrained environments, such as mobile robots with microcomputers. It is achieved by separating object-aware map building process. The proposed mapping process is divided into 3 stages, data acquisition stage, object-aware map building stage, and inference stage. Experiments are conducted with Turtlebot4 mobile robot with Raspberry Pi microcomputers, to validate its performance in low-powered devices. The result showed 66% of success rate in overall text list, including 60% success rate with input texts with description. This study contributes to mobile robotics by showing that even microcomputers with limited processing power can support object-aware navigation tasks through optimized mapping method.
Indirect Time-of-Flight (iToF) technology is widely utilized for precise spatial sensing. However, it is highly suscep-tible to Multi-Path Interference (MPI), which can be induced by semi-transparent objects, sharp edges, or fog and degrade the accuracy of 3D measurement. This study proposes a hybrid optimization approach combining Orthogonal Matching Pursuit (OMP) and Sequential Quadratic Programming (SQP) to effec-tively mitigate MPI in iToF systems. By using OMP to generate initial values for SQP optimization, we demonstrate improved accuracy and faster convergence compared to traditional meth-ods like Particle Swarm Optimization (PSO) combined with the Interior Point Method (IPM). We evaluate the proposed OMP-SQP hybrid method using a publicly available dataset, which includes various challenging scenarios such as stray light interference and closely spaced targets. The experimental results show that the OMP-SQP method achieves up to 3.33 times faster optimization with comparable or even superior accuracy to the existing method. These findings suggest that the OMP-SQP hybrid method is a highly efficient and reliable solution for enhancing the performance of iToF systems under complex MPI conditions.
Sentiment analysis on opinions, views, emotions, or attitudes from written statements, comments, etc. is a research topic of growing interest. This work can benefit service providers, corporations, governments, and individuals in gathering information and making opinion-based decisions. At educational institutions, especially higher education institutions, periodically collecting student feedback about teaching activities is extremely important in adjusting teaching methods. Our research is aimed at determining a hybrid deep learning model applied to the problem of analyzing emotions from text that can be used for the task of analyzing student feedback.
As a class of exact Bayesian filtering algorithms for non-linear/non-Gaussian recursive estimation of dynamic stochastic systems which can randomly switch on and off, the Bernoulli filter has been extensively studied and applied in target tracking and other dynamic phenomena. In general, there is no analytic solution for the Bernoulli filter, and it is implemented in two ways: Monte Carlo approximation and Gaussian-sum filter (GSF) based on Gaussian-sum models. GSF can be given in the analytic form and has the computational advantage. The possibility Bernoulli filter, which is based on Uncertain Finite Set (UFS) instead of Random Finite Sets (RFS) and implemented as an analogue of standard Bernoulli filter, has been introduced to address the epistemic uncertainties arising from imprecise or partial knowledge of models and/or filtering parameters. Similarly, the Bernoulli Gaussian-max filter (GMF) is formulated as an analogue of the Bernoulli GSF in the framework of possibility theory, rather than the framework of probability theory, with an objective to achieve enhanced ro-bustness. The performance of the Bernoulli GMF and Bernoulli GSF is evaluated and compared through simulation tests in a single target tracking application.
Nowadays, big data in livestream (BDL) is becoming increasingly prominent in computing systems (CS). Algebraic structures apply algebraic concepts to BDL in CS, aiding in the analysis and resolution of BDL-related problems. This research focuses on algebraic structures related to BDL, such as monoids, to analyze and categorize their properties. These algebraic monoids are essential in formal methods for verifying and analyzing applications of BDL, as they model program behavior and ensure program correctness. This paper examines the algebraic structure of BDL in CS, specifically detailing BDL and formalizing the properties of BDL monoids. Generally, algebraic structures are crucial for computing BDL, offering fundamental frameworks for comprehending and addressing complex issues of BDL in CS.
This paper presents the design and optimization of deep neural networks (DNNs) tailored for multi-label classification on smart electrical devices monitoring system, an emerging category of intelligent measurement devices. The smart system is equipped with advanced sensing and data acquisition capabilities, generating complex datasets necessitating sophisticated analytical methods. This complex dataset comprises various properties regarding electrical devices, which can be used to train and evaluate an artificial intelligence (AI) model for classifying home devices when plugged in. Various DNN architectures are explored by using neural architecture search (NAS) and optimization techniques like pruning to enhance classification accuracy and computational efficiency. The article's approach includes leveraging transfer learning, regularization methods, and hyperparameter tuning to address the challenges of multi-label classification. Experimental results demonstrate significant improvements in prediction performance compared to traditional methods. The findings underscore the potential of optimized DNNs in advancing the functionality and reliability of the smart system in diverse application scenarios.
