This paper addresses the problem of estimating the poses of multiple objects under sparse and noisy data in robotics systems in unknown environments. Due to the data sparsity caused by limited data acquisition policies for saving resources and the data noises caused by environmental disturbance, the robustness of the conventional methods for the estimation degenerates. This paper proposes a robust multiobject pose estimation method to solve the problem, which can be applicable to not only ground environments but also space environments. First, a robot builds a map as a model its surrounding environments by simultaneous localization and mapping (SLAM). Then, multiple objects are detected by a visual sensor with a deep learning-based algorithm. Finally, the poses of multiple objects are robustly estimated by probabilistic filters with range data and localize them in the built map. The performance of the proposed method was tested with a real robotic system with visual and range sensors. Experimental results showed that the proposed method can robustly estimate the poses of multiple objects under sparse and noisy data.
Packet classification is the function that eliminates or forwards arriving packets at a network device according to packet classification policies. Packet classification using rule lists is widely used because of its low memory requirements and ease of implementation. However, as the number of rules increases, the classification time increases, and the packet communication latency increases. In this paper, we propose a method that rewrites rules to eliminate the precedence constraint and places the rule with high matching frequency at a higher position. This method of expanding and restructuring rules, which makes rule replacement easier, can reduce latency quickly without disrupting network equipment operation. Computer experiments using ClassBench [1], a benchmark for packet classification algorithm, confirm the effectiveness of the proposed method.
This paper presents a comprehensive comparison between Reduced Capability (RedCap) and 5G New Radio (NR). RedCap is the first standardized 5G technology for the Internet of Things (IOT). RedCap devices were introduced in third Generation Partnership Project (3GPP) Release 17 as a NR light with the target of cost-efficiency and energy saving using narrower bandwidth and lower modulation complexity. Redcap bridges the Narrow Band IOT and 5G NR. Reducing complexity in modulation, bandwidth, antenna configuration, and modem capabilities, significantly reduced the cost and energy consumption. RedCap and 5G NR broadband devices in terms of the specification, throughput and architectures focused on power consumption in band n78 are compared. The reduced capabilities in RedCap devices make them simplified front end modules with lower complexity and power consumption. The results highlight design trade-offs relevant to developing IOT and consumer RedCap devices in mid-band spectrum. Scattering parameters, gain, and throughput are presented by Advanced Designed System (ADS) and MATLAB simulation. The results confirm that RedCap devices achieve sufficient throughput and energy efficiency for a wide range of low data rate use cases, such as smartwatches and industrial sensors.
This paper presents a novel Ultra-Wideband Orthogonal Frequency Division Multiplexing with Index Modulation (UWB-OFDM-IM) system, which combines the spectral efficiency of OFDM-IM with the robustness of UWB in multipath environments. The proposed system divides subcarriers into groups, activating only a subset to transmit QPSK-modulated symbols while leveraging subcarrier indices to carry additional information bits. The simulation results of the proposed technique compared with conventional UWB-OFDM. The results are as follows: 8.5 dB SNR gain at a BER of 10-3, a 1.8 dB reduction in PAPR, and a 6.02 dB improvement in spectral containment at 5 GHz. These improvements make UWB-OFDM-IM suitable for energy-constrained IoT applications, offering reliable connectivity in dense and multi-path environments.
Authentication is a critical security component in both cloud and IoT environments, and Keystone offers multiple methods to perform various deployment scenarios and security requirements. This paper presents a comprehensive analysis of authentication mechanisms available in OpenStack's identity service, Keystone. It systematically examines four key authentication methods: Password Authentication, Time-based One-Time Passwords (TOTP), OAuth Authentication, and Application Credential Authentication. Through hands-on demonstrations and comparative analysis, we evaluated the strengths and limitations of each method. The findings aim to guide cloud and IoT administrators and architects, specifically in 5G networks, in selecting and implementing the most suitable authentication methods for their OpenStack deployments based on their specific security requirements, integration needs, and operational constraints.
Smart home devices promise independence and convenience, yet their interfaces remain inaccessible to many older adults, people with disabilities, and non-technical users. This paper presents SmartHomeHarmonizer, a privacy-preserving conversational framework that unifies commercial voice assistants with Generative AI (GenAI) to enable natural, inclusive smart home control. We extend prior work by: (1) detailing a reproducible edge-first architecture with comprehensive implementation details, (2) reporting statistical significance, confidence intervals, and effect sizes for a 287-participant study across diverse populations, and (3) expanding security and threat modeling with formal privacy guarantees. Compared with leading baselines (Google Nest Hub, Amazon Echo Show), SmartHomeHarmonizer achieves 92% +/- 5% task success (vs. 77-78% for baselines), 25.3s median completion time (vs. 31.8-32.9s), and 4.5/5 satisfaction rating (vs. 3.8/5)-while keeping all data local. The system's modular design runs on affordable hardware (Raspberry Pi 5) and supports multiple device protocols (Zigbee, Z-Wave, Wi-Fi, Bluetooth, Matter), advancing digital inclusion without sacrificing privacy.
Public auditing (PA) enables the remote auditing of cloud data integrity, eliminating the need for downloading or directly accessing the data. However, existing PA schemes are vulnerable to quantum attacks because they are designed based on conventional mathematical problems. To resist such quantum attacks, quantum-secure PA schemes have been proposed based on lattice-based cryptography. Although these schemes resist attacks, they are inefficient and unsuitable for cloud-assisted IoT devices because they rely on the trapdoors, which create huge computational overhead. In this paper, we propose an Efficient and Quantum-Resistant Public Auditing Scheme (EQRPAS) for Cloud-Assisted IoT without lattice trapdoors to reduce computational overhead while resisting quantum attacks. The security of EQRPAS is proven under the Small Integer Solution (SIS) assumption. Additionally, the analysis of performance and implementation outcomes demonstrates that our approach is efficient, making it particularly suitable for resource-constrained devices.
This paper investigates Multi-Armed Bandit (MAB) algorithms for adapting the transmission parameters of a 2.4 GHz Long Range (LoRa) Internet of Things (IoT) network. The aim is to dynamically select the optimal transmission parameters to reduce energy use while ensuring reliable data transmission under dynamic operating conditions. For this, the most well-known MAB strategies: Upper Confidence Bound (UCB), Exponential weights for exploration and exploitation (EXP3), Epsilon-Greedy, Thompson Sampling (TS), and Tugof-War (ToW), are thoroughly evaluated using a reward function that balances energy efficiency, data reliability, and data rate. Our simulation results indicate that UCB consistently achieves the best compromise between computational complexity and transmission performance in all considered dynamic LoRa IoT settings, including stochastic and nonstationary ones.
Authentication is a key aspect of cybersecurity, ensuring that only legitimate users and devices access protected resources. Yet, many common protocols such as PAP, CHAP, RADIUS, and LDAP still suffer from critical weaknesses. This paper introduces a risk-based classification model using CVE, CVSS, and CWE data to assess their security level. Results show recurring flaws, including unencrypted credentials, weak cryptography, and replay or impersonation risks. We also outline current improvements, such as stronger encryption, biometric methods, and AI-driven detection. The study helps guide the choice of safer authentication mechanisms and supports the development of more resilient security infrastructures.
Identity and Access Management (IAM) in the Internet of Medical Things (IoMT) faces critical challenges, including the detection of sophisticated attack patterns and the execution of real-time threat responses. This work presents the IoMT Sentinel Platform, an adaptive security framework designed for IoMT environments. The platform integrates a finetuned DistilBERT model for anomaly detection in contextualized IAM and system logs, achieving an accuracy of 99.98 % on a simulated IoMT dataset. Beyond detection, the platform introduces a dynamic trust scoring engine that evaluates each device's security posture based on access history, network behavior, and anomaly records. This trust score directly influences an automated response engine that leverages Keycloak APIs to revoke tokens, adjust privileges, and block malicious IP addresses in real time. Experimental evaluations across multiple scenarios demonstrate the system's ability to detect volumetric and distributed attacks, enforce dynamic trust-based policies, and adaptively mitigate threats with low response latency. This work is among the first to integrate a lightweight language model, a dynamic trust scoring engine, and an autonomous IAM mitigation system into a unified IoMT security framework. Unlike previous approaches that focus solely on detection, the proposed platform emphasizes real-time defense through seamless coupling of anomaly detection and IAM policy enforcement. Our results demonstrate that such a synergy enables proactive and scalable protection in medical IoT environments.
The proliferation of the Internet of Things (IoT) ecosystem has enabled unprecedented data connectivity while increasing exposure to sophisticated cyber threats. Existing intrusion detection systems (IDS) struggle in IoT environments due to limited computational capacity and scarcely labelled malicious data. This paper introduces a novel hybrid meta-deep learning framework for effectively classifying malicious IoT packet data. Our approach integrates a Siamese Network for binary detection of threats and a Prototypical Network for multi-class classification of attack types. The model performs well on constrained datasets with minimal samples, using TON_IoT and CICIDS2018 datasets for empirical validation. Our results show a high detection accuracy of over 94%, proving its potential for robust, scalable intrusion detection efficiency and effectiveness in IoT applications. Our contribution is a two-stage few-shot IDS that pairs a Siamese similarity gate with a prototype-based attack classifier designed for edge constraints. Across CICIDS2018 and TON_IoT we achieve macro-F1 similar to 0.93 with strong performance on frequent classes, while also reporting the limits on low-frequency categories such as Infiltration and SQL injection and how we mitigate them under imbalance. We situate the approach alongside recent Transformer-based IDS trends and clarify where our design is complementary rather than competitive.
One promising way to enable distributed intelligence at the edge while protecting data privacy is to integrate Federated Learning (FL) with Unmanned Aerial Vehicles (UAV) networks. Using FL enables each UAV to cooperatively train a global model without sharing raw data, especially in UAV swarms used for surveillance, monitoring, or emergency response missions. But choosing the best clients (UAVs) for every training cycle is made extremely difficult by the dynamic and diverse character of UAV environments. These difficulties are brought on by things like fluctuating connectivity, shifting patterns of movement, and energy limitations. In this work, we investigate the problem of client selection for Federated Edge Learning in UAV networks. We first present a taxonomy of existing selection strategies, considering criteria such as model performance and UAV mobility. Then, we propose SafeLearning, an adaptive client selection framework that integrates both mobility-awareness and distance with speed to enhance learning efficiency and model accuracy. Simulations demonstrate that our method significantly improves convergence speed and reduces client dropout, while maintaining high model performance in dynamic UAV scenarios.
This work focuses on an experimental understanding of the hardware footprint and achievable performance of modern, container-native edge devices built on open-source platforms. As a case study, StarlingX was chosen because it can provide a Container-as-a-Service (CaaS) explicitly designed to meet low-latency and small-footprint requirements. Notably, full-fledged Infrastructure-as-a-Service (IaaS) can be provisioned concurrently at StarlingX edges. Still, such services are now built on top of containers, that is, IaaS over Container-as-aService (CaaS). However, the hardware demands of such versatile container-based architectures and their achievable performance have yet to be investigated. The analysis presented here considers the perspectives of both providers and clients. It brings results in CPU, memory, and I/O on-edge hardware for a direct comparison between virtual machines and container provisioning. It also devises an experiment to investigate whether CaaS suits multitenant services, such as network slicing at edges, where resources are limited and isolation is not as tight as in IaaS.
As a typical branch of distributed computing, edge computing plays an important role by shifting the computation unit from centralized servers to the local network end, thereby reducing resource consumption and alleviating network load. It allows data to be processed locally upon acquisition, without relying on unstable or limited infrastructure, making it an ideal solution for rural and remote areas, particularly in agro-forestry scenarios. Furthermore, AI and machine learning (AI/ML) techniques offer promising capabilities for accurate identification and prediction. To achieve lower redundancy, enable on-demand data analytics and optimal decision-making, this paper focuses on two key areas: IoT data processing/filtering and imagery-based object recognition/counting, with the purpose of digitalization and automation in the field. The performance results prove that the proposed edge solutions effectively reduce unnecessary data volume, achieving a false positive rate of only 1.03 % in data filtering tests and an 82 % decrease in imagery detection tests. These solutions capture only the relevant information or objects, enabling accurate and in-depth analysis based on user needs.
The growing demand for energy-efficient industrial systems is accelerating the shift from traditional SCADA architectures to more flexible and scalable designs. This study presents a modular open-source SCADA framework that integrates Arduino Uno R4 Wi-Fi, MQTT, Node-RED, and InfluxDB with edge computing on Raspberry Pi. Real-time monitoring is enhanced with predictive analytics, where a Random Forest model forecasts energy consumption from process and environmental temperatures. The model achieved high accuracy (MAE $=0.0042$, RMSE $=0.0068$) and was deployed via a Streamlit interface for accessible real-time prediction. The system demonstrated reliable sensor accuracy, sub-second actuation latency, and adaptive control informed by external weather data. This cost-effective framework provides a replicable pathway for SMEs to modernise legacy systems. It further demonstrates how edge computing, open-source IIoT platforms, and AI-driven control strategies can be integrated for smart manufacturing and energy-aware operations.
Applications deployed in resource clusters often experience performance variability due to resource contention and system heterogeneity. This is particularly problematic for time-sensitive workloads, such as latency-critical IoT applications, which are often deployed in edge environments where co-located services compete for limited computational resources. Existing scheduling policies often overlook runtime performance fluctuations and server heterogeneity, which can result in unpredictable task execution times. To address this, we propose a methodology for predicting both the mean and standard deviation of application round-trip time (RTT) under dynamic co-location scenarios. These lightweight variability predictors are deployed alongside applications and use historical RTT data to model application performance. We evaluate our approach using a real-world electron microscopy workflow with diverse resource requirements. Our predictors achieve prediction accuracy up to 94% while maintaining a low overhead under dynamic and heterogeneous conditions. Although the developed predictors are not yet integrated into the scheduler, simulation results demonstrate that they enhance application performance and reduce resource waste by enabling more informed allocation decisions.
This study proposes a web-based, real-time monitoring system for elderly individuals, leveraging low-cost, nonwearable ambient sensors to support safe and independent living at home. The system integrates Passive Infrared (PIR) sensors, temperature-humidity-barometric pressure sensors, CO2 sensors, and door proximity sensors. These devices transmit timestamped data to an edge computing unit, which performs local processing and synchronizes with a cloud server. The collected data are visualized through a browser-accessible interface, enabling caregivers and family members to remotely observe patterns of physical activity and indoor environmental conditions without requiring specialized applications. While a previously developed spatiotemporal data cleaning method is employed to enhance data consistency, the primary contribution of this study lies in the practical integration of sensing, processing, and visualization components into a cohesive, scalable framework. Pilot experiments in a mock residential setting confirm the system's feasibility and usability, demonstrating that meaningful behavioral and environmental insights can be derived from simple, unobtrusive sensors. This framework supports privacyaware, community-based elderly care and contributes to the development of accessible gerontechnology solutions for agingin-place.
This paper investigates the impact of Denial-of-Service (DoS) attacks on sampled-data control systems with non-uniform sampling and extends the analysis to cloud-based control architectures subject to random delays and packet losses. A Lyapunov-based framework is developed to ensure exponential stability of linear time-invariant (LTI) systems under DoS conditions, with sufficient conditions derived via linear matrix inequalities (LMIs). Numerical simulations illustrate the performance of both local and cloud-based controllers, revealing that while both architectures achieve stability, the cloud-based implementation suffers from degraded transient performance due to communication uncertainties. The results highlight the trade-offs between resilience and performance in modern networked control systems.
This paper introduces a novel integration of Large Language Models (LLMs) into website localisation services through a prompt-engineering-first approach. Traditional localisation methodologies have established trade-offs between quality, cost, and efficiency. The research explores how focusing on prompt engineering rather than post-translation editing can fundamentally transform localisation workflows while maintaining highquality outputs. A practical case study involving the localisation of a cross-platform application demonstrates superior quality compared to specialised translation services while offering significant advantages in maintaining marketing tone and reducing implementation complexity. The findings indicate that for projects with marketing-focused content and moderate translation volume, the LLM approach with focused prompt engineering represents a superior solution compared to traditional translation services.
A new approach is introduced to provide multicast delivery services in wireless networks and IoT deployments without the need to deploy any multicast routing protocol. The approach consists of allowing Internet Group Management Protocol (IGMP) messages to traverse multiple hops to allow routers to express their interest in receiving traffic from multicast groups. The proposed variant of IGMP with multi-hop interests that we call IGMM (for Internet Group Message Multicast) replaces the role of a multicast routing protocol and builds multicast forwarding tables used in the data plane for multicast traffic in much the same way that some information-centric networking (ICN) proposals do.
