Mobile robots require effective and secure path planning, especially in complex and dynamic environments. Traditional algorithms such as A*, Dijkstra, and Rapidly-exploring Random Trees (RRT) offer dependable and mathematical solutions but face challenges regarding scalability, adaptability, and processing requirements in real-time applications. On the other hand, artificial intelligence (AI) techniques, such as reinforcement learning (RL) and neural networks (NNs) provide flexibility and quick decision-making but face challenges such as data dependency, optimal solution, and computational overhead. This review analyzes hybrid path planning methodologies that integrate traditional algorithms with AI techniques, utilizing the advantages of both to overcome their limitations. Hybrid approaches improve scalability, collision avoidance, and re-planning efficiency by integrating the accuracy and reliability of traditional techniques with the adaptability and learning capabilities of AI. This review categorizes and analyzes research to identify significant gaps and suggests future paths for enhancing hybrid path planning, offering insights for the development of more robust and intelligent navigation systems for mobile robots and autonomous platforms.
Artificial Intelligence algorithms are increasingly applied to tasks in Natural Language Processing, including document clustering. As these algorithms become increasingly complex (such as transformer-based embeddings, like BERT) and/or are of a ``black-box'' nature, such as Graph Spectral Clustering (GSC) algorithms, the demand for explaining the results of such algorithms is becoming increasingly urgent. In this paper, we propose a model-aware method to explain the results of GSC in the context of BERT-based embeddings. We present a novel theoretical methodology for explanation, based on the premise that document similarity in GSC is computed as cosine similarity of BERT embeddings of documents. We demonstrate the validity of this methodology by presenting strong GSC clustering results, restoring the human-made assignment of hashtags to tweets. We show that GSC based on BERT embeddings outperforms approaches using Term Vector Space and GloVe embeddings. Therefore, the resulting explanations are also expected to be of higher quality.
The paper presents a robotic system that assists people with reduced mobility in the activities of picking up and putting down objects out of reach. Human‐robot com‐munication is non‐verbal, using gestures that have been specifically selected for the robot’s use. Gestures are read out using an RGB‐D camera while the commands they ex‐press are executed online by a small UR3 cobot. The eval‐uation of the system has shown that it is useful and safe in the sense of the SUS and GQS respectively.
Cybercrimes especially encompasses crime against children, data breaches, and privacy violations, are increasing in frequency due to the quick development of technology, which emphasizes the necessity of complex systems to analyse and categorize these offenses. There are many opportunities to analyze cybercrime data using Machine Learning (ML) techniques because of its enormous accumulation. This study proposes a model that has the potential to automatically analyze text-based reported cybercrime complaints based on the features by use of Random Forest (RF) and Gradient Boosting (GB) algorithms. This model includes a Bag of Words (BoW) approach for feature engineering to analyze reported cyber crime and suggest relevant Indian IT Act sections, such as Section 66E deals with privacy protection, Section 43A for reported data breach, and Section 72A for disclosure of information, using Natural Language Processing (NLP) for feature extraction and classification. This strategy enhanced the law and enforcement process by timely and accurately categorizing crime. By automating cyber law and providing timely legal answers to various reported cybercrimes, especially those concerning privacy and data protection, the model improves the capabilities of cybercrime units and achieves high accuracy and precision in anticipating pertinent legal sections.
The use of Artificial Intelligence (AI) has significantly advanced emotion recognition within Human-Computer Interaction (HCI). This paper aims to develop a multimodal emotion detection system for educational and work environments using an enhanced AI machine vision system. The primary focus is on training and testing a multimodal AI model in Python, utilizing Convolutional Neural Networks (CNN). The results from the trained facial emotion AI model demonstrated substantial improvements. Training accuracy increased from 30.49% to 72.21%, while validation accuracy improved from 37.6% to 60.58%. Simultaneously, training loss decreased from 180.69% to 73.65%, and validation loss reduced from 172.97% to 107.53%. This CNN-based model can detect seven emotions: happy, sad, neutral, angry, fear, disgust, and surprise, using OpenCV. Furthermore, the ECG emotion AI model, also trained with CNN, successfully recognized patterns for the same seven emotions. When these two models are combined into a multimodal AI system, they can detect facial and ECG-based emotions simultaneously. This comprehensive approach allows for the detection of both visible and hidden emotions, such as stress or anxiety, which may not be easily discernible through facial expressions alone. The integration of these models into a multimodal AI system provides a more accurate and holistic understanding of human emotions, enhancing applications in educational and work settings. The improved detection capabilities can lead to better user experiences and more effective responses to emotional states, ultimately contributing to advancements in HCI.
Solar photovoltaic energy is gaining popularity in modern distribution networks due to its clean energy attributes. In order to maximize PV power generation conversion, the application of Maximum Power Point Tracking is essential. Henceforth, this work presents a novel hybrid MPPT approaches based on Cascaded Adaptive Neuro Fuzzy Inference System and Radial Basis Function Neural Network to achieve rapid and greatest PV power extraction while ensuring zero oscillation tracking with a Single-Ended Primary Inductor Converter. The SEPIC converter efficiently regulates the output voltage to match grid requirements while maintaining high power conversion efficiency. The Cascaded ANFIS and RBFNN are combined to enhance MPPT accuracy and robustness under varying environmental conditions. The cascaded architecture enables seamless transition between the two controllers, ensuring optimal performance across an extensive range of operating conditions. The MATLAB/Simulink is used for analyzing the efficacy of the proposed system and the proposed converter and MPPT approach is compared with existing topologies for proving the importance of the developed work. The outcomes demonstrate that the proposed SEPIC converter achieves reduced THD of 1.16% and the Cascaded ANFIS-RBFNN based MPPT approach attains higher tracking efficiency of 99.61% with rapid convergence speed and execution time compared to traditional techniques. Overall, this paper represent a promising direction towards achieving more efficient and sustainable PV-driven grid integration.
The rapid development of Integrated Energy Systems (IES), which unify diverse energy technologies such as electricity, heat, cooling, and gas, has heightened the importance of optimizing their operational modes. This paper explores the application of ternary optimization in IES, a discrete optimization approach where variables are constrained to three values: {-1, 0, +1}. Ternary optimization offers a balanced trade-off between binary and full-precision optimization, providing significant advantages in computational efficiency, memory savings, and energy efficiency. The article covers: key concepts of ternary optimization, including ternary representation, sparsity, and quantization; advantages and challenges of ternary optimization, such as reducing computational complexity and potential loss of accuracy; the application of ternary optimization for the IES. The role of ternary optimization in simplifying energy flow management, reducing computational resources, and enabling faster decision-making in dynamic environments is emphasized. Examples of using ternary optimization for energy distribution, microgrid management, integration of renewable energy sources, and energy storage systems are provided. A practical example of transforming an optimization model for IES into a ternary model using GMPL (GNU MathProg Language) is provided, demonstrating how ternary variables, constraints, and objective functions can be adapted. The paper concludes by discussing promising directions for ternary optimization in IES, including integration with AI and machine learning, development of specialized algorithms, and hardware support for ternary computations. Research underscores the potential of ternary optimization to enhance the efficiency, resilience, and scalability of IES, particularly in the context of increasing renewable energy integration and the complexity of modern energy grids.
Irreversible vision loss, which often develops slowly and with no outward signs of illness, is most commonly caused by glaucoma. Because it may slow the disease's progression, the initial stages of glaucoma detection are of the utmost importance. Ordinary procedures and manual assessments are based on traditional diagnostic techniques, which are notoriously imprecise. It follows that automated glaucoma analysis is critically important for the early and precise detection of glaucoma. Also, on the other hand, the medical image dataset is mostly imbalanced in nature. To overcome all these issues, the present research work developed an effective framework by Utilising Generative Adversarial Networks (GAN) to synthesize images to balance out the dataset. For example, when dealing with fundus images, conventional methods, such as translation from image-to-image operations, are used. In particular, these techniques are employed to produce synthetic fundus images and the associated vessel networks. Improving the synthetic images' quality as a whole and capturing finer details is the main goal. The goal of this effort is to improve synthetic fundus images in terms of accuracy and authenticity, which will lead to new developments in the area of fundus image synthesis. Initially, a raw dataset has been preprocessed using the Gaussian filtering technique, which helps to minimize the unnecessary noise over the images. Then GAN is used to balance out the dataset, which helps to produce synthetic images and produces reliable outcomes in classification tasks. The next segmenting optic cup is done using the Enhanced Level Set Algorithm. Finally, Pretrained MobileNetV2 is used for the accurate classification of glaucomatous images into normal and abnormal. Experimental results show that our proposed frameworks perform well compared to existing approaches with an accuracy of 98.9%.
T-way combinatorial testing is an essential approach for optimizing test suite generation by systematically covering parameter interactions while minimizing test cases. Various metaheuristic strategies have been introduced to improve test suite generation, with an increasing focus on balancing exploration and exploitation for efficient test selection. This study investigates the Sand Cat Swarm Optimization (SCSO) algorithm as a metaheuristic strategy for t-way test suite generation. Inspired by the hunting behavior of sand cats, SCSO dynamically adjusts sensitivity factors to improve test suite generation efficiency. To evaluate SCSO’s performance, 30 benchmark experiments were conducted across four groups of t-way configurations with t varying from 2 to 6 and v ranging from 2 to 10. Each configuration was executed five times, and the smallest test suite size was selected for analysis. Experimental results demonstrate that SCSO outperforms 15.79% of competing strategies, achieves comparable performance in 42.11% of cases, and is outperformed in 42.11% of benchmark comparisons. These findings highlight SCSO’s capability to generate competitive test suites, particularly in t-way interaction testing. The statistical evaluations, including Wilcoxon Rank and Friedman Mean Rank tests, further validate SCSO’s performance in comparison to other metaheuristic approaches. Although SCSO effectively reduces test suite size while maintaining interaction coverage, further enhancements are necessary to improve its adaptability and computational efficiency across diverse configurations. Future work should focus on refining SCSO’s exploration mechanisms to optimize search efficiency and extend its applicability in combinatorial test generation.
Electrification in the transportation sector, such as electric two-wheelers (E2W), including electric bicycles and motorcycles, must be accompanied by the development of a sufficient ecosystem. Meanwhile, currently, available commercial E2W battery chargers cannot keep up with user mobility in terms of charging time, which implicates the inflexibility of E2W in daily usage and needs to meet safety standards according to the battery specification. This work proposed a charging system to meet the need for fast and safe charging in conventional E2W batteries. It used an interleaved buck converter (IBC) regulated by a combination of proportional integral–fuzzy logic control (PI-FLC) algorithms to ensure safe and fast charging. This system is adaptable and potentially meets the universal principle by allowing users to adjust current and voltage based on the power source and vehicle specifications. The PI-FLC algorithm determines the optimal charging current by considering charging voltage, current, and state of charge (SoC). The IBC’s ripple cancellation feature enhances compactness by reducing the output filter capacity. The system uses nickel manganese cobalt (NMC) batteries with a specification of 72 V at 20 Ah with a 1 C charging rate. The performance was compared to PID CC-CV and PI-CV algorithms. The results showed that the proposed system produces a suitable voltage and current charging characteristic and only needs 57.75 minutes to charge 0-100% SoC, while PID CC-CV takes 180 minutes and PI CV takes 240 minutes.
The purpose of this study is to develop and evaluate ARIS (A Real-time Interactive Social Robot), based on a Turtle 4 platform aimed at improving human-robot interactions (HRI) on university campuses. ARIS combines a 3D printed social robot structure with a software architecture based on 2D LiDAR, odometry, and IMU sensors for navigation and mapping, in addition to a voice assistant structured in 3 stages: audio recording, transcriptional processing, reaction, and reproductive synthesis. Experimental results show that the success rate of ARIS is greater than 86.5%, maintaining high accuracy in navigation and obstacle avoidance. The system also offers performance consisting of voice interaction with a total reaction latency of 1500 to 2200 ms. Access to low-cost robotic platforms allows students and researchers access to practical training, customization, and validation in the development of new technologies in social robotics.
In recommender systems, collaborative filtering (CF) is a crucial technique, but it often struggles with the data sparsity issue, which impact the recommendation accuracy. To address this challenge we have proposed a Co-training Ensemble Learning (CTEL) technique that integrates item-based Collaborative Filtering, user-based collaborative filtering (CF), and Singular Value Decomposition (SVD) through a structured stacking methodology to improve the recommendation performance. The co-training procedure, which creates pseudo-labels for unlabeled data based on a confidence threshold, is used to iteratively improve the user-based and item-based CF models after they have been originally trained. These models produce predictions for validation and test sets, in conjunction with the independently trained SVD model. These forecasts yield meta-features, which include extra statistical variables like variance and product of predictions. The Linear Regression model is trained as the meta-learner to find the prediction of the base models in the best possible way using K-Fold cross-validation. Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE) are used to assess the final model's performance on a test set. The outcomes confirm the effectiveness of the co-training and stacking strategy by showing notable increases in prediction accuracy when compared with existing techniques. Using the advantages of collaborative filtering approaches and matrix-based approach, the proposed model offers a comprehensive foundation for creating advanced recommendation systems.
For an important segment of the Indian people, agricul‐ ture serves as a primary source of income. Most Indian farmers choose to produce crops in a field using tradi‐ tional farmingmethods; hence, one of their biggest issues is that they frequently choose to cultivate the incorrect crop for their soil type. The crop recommendation system proposed in this research would assist farmers and edu‐ cate them on decision‐making regarding which crops to plant on their property. Using soil parameters like potas‐ sium, nitrogen, and phosphorus as well as environmental variables like humidity, rainfall, and pH levels, to build this recommendation system, we used ML methods such as Random Forest, KNN, Naïve Bayes, SVM, and Logistic Regression. As a result, we also present comparative per‐ formance on the model for the dataset. Therefore, finally, these technologies will be helpful for farming and agri‐ culture. Today’s smart agricultural solutions, can address the growing concern about the world population’s food consumption and environmental impact. The accuracy of this crop recommendation system will depend on the following: The quality and quantity of our dataset, the relevance and effectiveness of our features, the choice and tuning of our machine learning models, the balance of our dataset and the complexity of the crop prediction task, performing thorough training, validation, and test‐ ing will give the accuracy metric we need.
Facial recognition technology finds applications in security, surveillance, and social media. Existing research explores the use of machine learning and deep learning for face recognition, emphasizing the need for improved accuracy. This paper proposes a system for suspect identification using facial recognition. The system leverages ensemble learning, combining machine learning and deep learning algorithms. It integrates seamlessly with OpenAI's advanced technologies and is supported by a robust cloud infrastructure. The proposed ensemble model's performance is compared to individual models like VGG-Face, Facenet, Facenet512, Deepface, DeepID, ArcFace, and SFace. The comparison uses multiple detectors and the Labelled Faces in the Wild (LFW) dataset. The results show that the ensemble model offers the most efficient processing time across all sample sizes. In contrast, models like VGG-Face and DeepID exhibit a steeper increase in processing time, suggesting lower scalability. For instance, at a sample size of 50, the local test completes in 61.3 seconds, while the cloud API test takes 67.2 seconds. This highlights the faster processing speed of the local test across all sample sizes..
The integration of generative models into robotics has marked a significant paradigm shift, promising to enhance the capabilities of robotic systems and expand their application across various sectors. This survey study explores the impact of generative models on robotic innovation, delving into the conceptual and technical advancements they have spurred, the broad spectrum of their applications, and the challenges and future directions they present. Through a comprehensive literature review and analysis, this study highlights how generative models have driven advancements in robotic perception, learning, and decision-making capabilities. Applications in sectors such as manufacturing, healthcare, autonomous vehicles, environmental monitoring, and agriculture underscore the transformative potential of these technologies. However, the integration of generative models into robotics is not devoid of challenges, including technical limitations, ethical concerns, and societal implications. The study concludes by envisioning future directions that focus on enhancing model efficiency, addressing data bias, improving interpretability, and fostering interdisciplinary collaborations. By navigating these challenges, the continued evolution of generative models in robotics holds the promise of unlocking new levels of innovation and societal benefit.
Modern IP networks face significant challenges in maintaining performance under dynamic and diverse traffic conditions. Traditional congestion control algorithms, such as TCP Reno, Cubic, and even recent reinforcement learning (RL) methods like PPO and DQN, often respond uniformly to packet loss, failing to distinguish between congestion-induced losses and those arising from wireless interference or hardware failures. This paper introduces AHMA (Adaptive Hierarchical Meta-Agent) — a novel, two-stage intelligent congestion control framework that integrates a Bayesian Transformer-based classifier with a Meta-Evolutionary Reinforcement Learning (Meta-ES-RL) controller. AHMA first classifies the cause of packet loss in real-time and then dynamically selects an optimized control strategy based on classification confidence. Using a synthetically generated NS-3 dataset of 1000 labeled flow samples, we evaluate AHMA’s performance against PPO, DQN, TCP Cubic, and TCP Reno across key metrics. Experimental results show that AHMA achieves a decision accuracy of 92%, reduces packet loss to 8.56% with improved throughput, and decreased latency, outperforming all baseline methods. This approach represents a significant advancement in adaptive, cause-aware congestion management, with strong potential for deployment in next-generation high-performance IP and 5G networks.
Efficient and safe navigation of Unmanned Aerial Vehicles (UAVs) is critical for various applications, including combat support, package delivery and Search and Rescue Operations. This paper introduces the Tangent Intersection Guidance (TIG) algorithm, an advanced approach for UAV path planning in both static and dynamic environments. The algorithm uses the elliptic tangent intersection method to generate feasible paths. It generates two sub-paths for each threat, selects the optimal route based on a heuristic rule, and iteratively refines the path until the target is reached. Considering the UAV kinematic and dynamic constraints, a modified smoothing technique based on quadratic Bézier curves is adopted to generate a smooth and efficient route. Experimental results show that the TIG algorithm can generate the shortest path in less time, starting from 0.01 seconds, with fewer turning angles compared to A*, PRM, RRT*, Tangent Graph, and Static APPATT algorithms in static environments. Furthermore, in completely unknown and partially known environments, TIG demonstrates efficient real-time path planning capabilities for collision avoidance, outperforming APF and Dynamic APPATT algorithms.
3D Object Localization has been emerging as one of the main challenges in Machine Vision tasks. In this paper, we proposed a novel 3D object localization method, leveraging a blend of deep learning techniques primarily rooted in object detection, post-image processing, and pose estimation algorithms. Our approach involves 3D calibration methods tailored for low-cost industrial robotics systems, requiring only a single 2D image input. Initially, object detection is performed using the You Only Look Once (YOLO) model, followed by an R-CNN model for segmenting the object into two distinct parts, i.e., the top face and the remainder. Subsequently, the center of the top face is served as an initialization position, and being refined with a novel calibration algorithm. Experimental results demonstrate a notable reduction in localization error by 87.65% when compared to existing methodologies.
In health care, there is a growing interest on building recommendation systems for sleep apnea management. These systems use data from a variety of sources, including patient-reported outcomes and electronic health records, to assess sleep quality, breathing patterns, and medical treatment adherence. Leveraging artificial intelligence (AI), machine learning (ML), the Internet of Things (IoT), and cloud platforms, the system analyzes these data to uncover patterns and correlations. It then creates individualized patient profiles that incorporate details about diet, medical history, and sleep habits. Based on these profiles, customized recommendations are generated to enhance sleep apnea management. These recommendations may encompass treatment options and lifestyle adjustments, Yoga, exercise, etc. to improve treatment effectiveness and overall well-being for individuals with sleep apnea. This review article discusses available literature on sleep apnea, its diagnosis, and the role played by ML and deep learning classifiers in the prediction and classification of the disease. The article also presents a comparative analysis on performance measures for these methods. This article highlights the research scope for incorporating technologies such as AI, the IoT, and computational intelligence in improving the diagnosis, remote monitoring, and treatment of sleep apnea.
The SARS‐CoV‐2 pandemic has heightened the need for advanced and automated disinfection methods to ensure workplace safety and hygiene. This study presents the de‐ sign and implementation of a robotic vehicle capable of autonomously disinfecting high‐risk areas in diverse work environments based on human activity levels. The sys‐ tem integrates a machine vision module using YOLOv5 for real‐time human detection, a Decision Support System to assess contamination risks, an autonomous navigation module for path planning, and a user‐friendly interface for operator control. By leveraging real‐time data, the robot precisely applies disinfectant to identified high‐risk zones, dynamically adjusting the spray volume based on the level of contamination. The system was validated in a real‐world workplace setting, demonstrating its ability to autonomously perform targeted disinfection, offering a scalable solution to support workplace hygiene.
