Traffic accidents cause 1.3 million deaths annually worldwide, with 93% attributed to human error. This paper presents Edge data for Safety (E4S), a Smart Mobility live pilot deployed within the ToMove Living Lab of the city of Turin. The E4S infrastructure includes a pioneering edge computing server featuring secure data processing and interoperable API exposure, compliant with ETSI Multi-access Edge Computing (MEC) standards. It interacts with the ToMove digital twin platform, which integrates data from 15,000+ connected vehicles, 527 roadside cameras, 66 traffic sensors, and the 5T public transport network. Our open and secure multi-tenant edge server processes authenticated API calls with low latency, featuring an advanced API Gateway, granular access policies, and GDPR-compliant data management. The system serves both municipal services and third-party commercial operators (e.g. InsurTech and fleet management companies). Users are engaged through a mobile app, receiving Vulnerable Road User safety alerts and In-Vehicle Information notifications. Experimental results indicate that open multi-tenant edge platforms can successfully balance innovation, commercial exploitation, and regulatory compliance, offering a scalable blueprint for other Smart Cities as an example of impactful solution for the emerging consumer-technology ecosystem.
Preventive screening is moving into consumer settings where devices must deliver timely feedback under limited compute, intermittent connectivity, and strict privacy expectations. This paper contributes a conceptual framework, reference architecture, and design guide for privacy-preserving edge artificial intelligence [AI] in consumer health devices, supported by scenario-based evaluation across home, kiosk, and remote deployments. The architecture integrates acquisition, local preprocessing, optimized inference, secure storage, selective synchronization, and interoperability to guide device, edge, and cloud partitioning. The analysis shows that guided acquisition, quality checks, first-pass risk scoring, and secure storage should remain near the point of use, while updates, fleet management, longitudinal analytics, and care-system integration can be externalized when policy and connectivity permit. Home systems prioritize guided usability and offline inference, kiosks benefit from stronger local edge resources and multi-user isolation, and remote systems require local autonomy with deferred synchronization.
Modern consumer devices demand real-time artificial intelligence (AI)-based inference under strict battery, latency, and computation constraints. Existing cloud-based inference introduces communication delay. Static edge inference and traditional device–edge collaboration do not address redundant computation among nearby consumer devices that observe similar contexts. This article proposes an energy-adaptive and sustainable edge AI framework for smart consumer devices. The novelty of our model lies in combining three mechanisms, namely, energy-aware model scaling, context-aware micro-federation, and adaptive inference-role coordination within an integrated consumer-edge workflow. The framework evaluates device-level inference cost and runtime suitability and selects lightweight or high-accuracy models accordingly. This allows nearby devices with overlapping contexts to reuse inference outputs from selected primary nodes. Evaluation using OPPORTUNITY and ExtraSensory context-recognition datasets shows that the proposed framework reduces normalized energy consumption by 39%. The model reduces latency from 210 ms to 82 ms and redundant computation by 52%, while maintaining 95.8% inference accuracy. These representative results indicate that the proposed framework improves real-time responsiveness, reduces unnecessary device computation, and supports sustainable edge intelligence for next-generation consumer electronics.
The smart home consumer electronics has resulted in substantially higher household energy consumption, necessitating the development of intelligent and energy-efficient management systems. To support the goals of green computing and sustainable smart home environments, it is important to be able to accurately predict energy usage and schedule appliances in an efficient way. This study introduces a novel Energy-Efficient DT-LightGBM Hybrid Framework (EcoDT-GBM) designed for the efficient prediction and optimization of household energy consumption. The proposed EcoDT-GBM framework uses DT-based structural learning and LightGBM-based boosting to make predictions that are both accurate and easy to compute. The proposed model attains a Root Mean Squared Error (RMSE) of 0.0364, Mean Squared Error (MSE) of 0.0013 and a the coefficient of determination (R2 ) of 0.9989. Also, the proposed energy optimization framework is able to save an estimated 45.41% of energy by moving peak loads intelligently.
Consumer cameras are greatly used as everyday sensing interfaces. However, developing robust AI for these consumer-facing scenarios remains challenging because high-quality labeled facial data are often scarce, imbalanced, weakly supervised, and demographically imperfect. This article presents a data-efficient AU-Graph transfer learning framework for consumer-camera pain-aware sensing. Instead of treating facial pain estimation as a purely clinical recognition task, the proposed framework transfers structured Action Unit (AU) knowledge from DISFA, leverages SynPAIN-based synthetic pretraining to introduce pain-related and demographic variability, and adapts the model to a limited target domain through auxiliary AU graph supervision. By representing AUs as graph nodes and modeling their relational dependencies, the framework provides interpretable intermediate representations for trustworthy consumer-facing perception under imperfect data. Under the fixed UNBC fold-1 protocol, DISFA-based AU-graph transfer improves macro-F1 over the corresponding backbones. SynPAIN is used to examine synthetic-source transferability and subgroup behavior.
The rapid expansion of massive Internet of things (IoT) deployments and the transition toward 5G networks have intensified the need for secure and scalable Remote SIM Provisioning (RSP) mechanisms. Unlike the Consumer and Machine-to-machine (M2M) RSP families, IoT-RSP operates across heterogeneous device classes, intermittent links, and delegated management roles, resulting in a broader attack surface and increased exposure to protocol-level threats. This article presents an integrated investigation of IoT-RSP that combines ProVerif-based symbolic analysis with reproducible benchmarking. The formal analysis shows that the dominant risks arise from binding, authorization, completion evidence, state consistency, and key-custody assumptions rather than from direct breaks of the profile-encryption primitive. The performance evaluation reports a physical consumer eUICC baseline of 50 successful profile installations with a median time of 27.991 s, an eIM-emulated ESipa control-plane benchmark under independent link impairment, local TLS/reuse/payload sensitivity, STM32F446RE/Mbed TLS crypto-kernel timings, and calibrated fleet-simulation results. These results clarify where security hardening remains practical, where this Cortex-M4 testbed exposes bottlenecks, and how IoT RSP should be deployed within the limits of the available testbed.
Consumer devices for healthcare, including smartwatches, patches, and mobile sensors, are reshaping post-surgical care applications through continuous at-home monitoring. In spite of more responsive and personalized healthcare that they can realize, wearable devices typically suffer from resource constraints (e.g., battery capacity and computational capability) and unreliable connectivity. On top of that, continuous transmissions of all sensed data may accelerate energy depletion and cause higher interference to other devices. Additionally, large-scale deployment of wearable healthcare devices may bring considerable privacy and security risks because of the frequent exposure of sensitive data. In this context, we propose a lightweight on-device intelligent scheduling framework for wearable healthcare systems. The proposed approach integrates deep reinforcement learning with adaptive wireless access, so that devices can selectively transmit sensing data based on system conditions such as data urgency, energy availability, and wireless channel. At the same time, non-critical sensors can conserve energy through radio frequency harvesting. By reducing unnecessary transmissions through context-aware decision-making, the proposed framework can improve energy efficiency and data freshness. We discuss key security and privacy challenges in artificial intelligence-enabled wearable systems, including inference-based attacks and data leakage risks. Finally, we discuss how lightweight on-device intelligence can improve communication efficiency, energy sustainability, and privacy in wearable healthcare monitoring systems
Despite the success of foundation models in language and vision, their expansion into embodied AI is bottlenecked by a lack of generalized touch sensing. This limitation is especially relevant to consumer electronics, where smartphones, wearables, VR controllers, home robots, and health monitoring devices require safe and adaptive physical interaction. Constrained by hardware heterogeneity and the necessity of active physical data collection, current haptic models remain rigidly task-specific. To overcome these limitations, this article explores the transformative potential and developmental trajectory of Haptic Foundation Models (HFMs). We detail the paradigm shift required to transition from passive Large Language Models and Vision Language Models into active HFMs across four core dimensions: action coupling, physical dynamical representation space, continuous time-series data granularity, and action-conditioned future state prediction. Furthermore, we synthesize existing large-scale tactile datasets and benchmark UniTouch, AnyTouch, T3, and Sparsh on TacBench for force estimation, slip detection, and relative pose estimation.
Connected and Automated Vehicles (CAVs) are increasingly recognized as consumer electronics devices with stringent cybersecurity requirements.Traditional Security Information and Event Management (SIEM) architectures are difficult to deploy in Connected and Automated Vehicles (CAVs) because they must operate under bandwidth constraints, limited on-board resources, and strict latency requirements. This challenge must be addressed to enable effective event aggregation, correlation, and fleet-wide monitoring without overloading the vehicle or delaying security response. In this context, we present a hybrid SIEM architecture for CAVs, combining an on-board lightweight SIEM augmented by remote edge SIEM. The architecture uses a security event distribution strategy that incorporates lightweight AI-based event prioritization directly within the vehicle to filter, rank, and forward only the most relevant security events. We validate a proof of concept implementation of our architecture and we show that, compared to traditional SIEM architectures, in-vehicle processing significantly reduces the number of transmitted events by up to 89% and improves the detection latency by up to four times, while preserving global monitoring and cross-vehicle correlation.
Effective sleep monitoring remains challenging in home settings due to the limitations of existing wearable devices and the intrusiveness of clinical-grade systems. This work presents a skin-conformal, pasteable platform for sleep monitoring and closed-loop posture correction. The system deploys three flexible modules attached to the forehead, neck, and torso, combining distributed inertial measurement units (IMUs) with application-specific acoustic and haptic components for multimodal sensing of upper-body posture and acoustic events associated with snoring and bruxism. A lightweight k-nearest neighbors (KNN) classifier implemented in a companion smartphone application performs real-time posture recognition and abnormal event tagging. When acoustic events are detected and associated with posture configurations, targeted vibrotactile feedback can be provided to prompt voluntary adjustment, forming a closed-loop intervention mechanism. The prototype evaluation in 10 participants and 33 distinct postures demonstrates that posture-classification accuracy exceeds 94% under controlled conditions. This proof-of-concept prototype demonstrates the feasibility of a privacy-preserving, extensible, and user-friendly framework for personalized in-home sleep monitoring.
Credible low-latency camera streams are essential for vehicle-in-the-loop (ViL) safety validation in edge–cloud vehicular IoT. Existing generation methods mainly address geometric alignment, with limited control and evaluation of photometric consistency. We propose Digital Twin Driven Video Stream Generation in ViL (DTDViL), which synchronizes a production vehicle with a virtual ego twin and injects controllable visual stimuli under native timing constraints. Photometric consistency is improved through sun-geometry-based shadow alignment and local Lab-space illumination and weather calibration. Experiments using YOLO11 show detector-response coefficients of variation below 2.6%. Target-level CIEDE2000 and boundary discontinuity further quantify appearance consistency and target–background fusion quality. Removing Lab-space calibration reduces mean confidence by 0.021–0.035 across four road-user categories. The average and maximum online critical-path latencies are 21.05ms and 32.07 ms, respectively, both within the 33.3ms budget for 30 FPS operation. These results support DTDViL as a credible real-time camera-stream generation framework for perception-side ViL evaluation.
Consumer health support is increasingly delivered through smartphones, wearable devices, home monitoring systems, and patient-facing digital platforms, yet many consumer-facing AI tools still provide generic or weakly bounded responses. This limitation is important for tasks such as symptom triage, medication awareness, wearable-data interpretation, and visit preparation, where useful support may require personal context, trusted resources, intermediate steps, and clear safety boundaries. Agentic AI offers a way to address this gap by coordinating safe multi-step work, tool use, and responses grounded in relevant information for context-dependent consumer health support. In this paper, we position agentic AI for consumer health relative to related work and propose a framework organized around consumer health inputs, safety and boundary control, agentic orchestration, trusted resources and tools, and bounded response generation. We then present a taxonomy of major application areas and illustrate the framework through a visit-preparation use case. We also report a prototype evaluation across sleep coaching, medication safety support, and visit preparation using synthetic scenarios. Finally, we discuss practical implications, deployment challenges, and future directions for safe and trustworthy consumer-facing agentic health systems.
Federated learning (FL) empowers privacy-aware consumer electronics but remains vulnerable to backdoor attacks that compromise safety-critical applications. We present SpecGuard, a server-side defense designed for communication-constrained consumer electronics and AIoT systems. SpecGuard characterizes client updates via a parameter-only sensitivity spectrum and filters malicious participants using the Wasserstein-1 distance to a benign prototype. Crucially, this approach requires no auxiliary data, trigger knowledge, client-side modification, or additional client communication. We evaluate SpecGuard on MNIST, CIFAR-10, and CIFAR-100 against representative adaptive attacks. Experiments using a MobileNet backbone, reduced-participation scenarios, dispersion-gate sensitivity analysis, anomaly-score visualization, and optimized server-side runtime measurement show that SpecGuard mitigates backdoors under the evaluated attacks while introducing no extra client communication and only limited server-side overhead.
Generative AI is garnering significant attention in industry and academia due to its remarkable ability to emulate human intelligence in decision-making and producing creative content, with applications across consumer health, consumer electronics, and consumer fashion. In this paper, we explore the applications and scope of generative AI in the consumer electronics space. To demonstrate their feasibility, we implement three real-world case studies: a GAI-based chatbot for consumer electronics queries, customised accessory generation for consumer electronics products using GAI, and GAI-based virtual try-ons for fashion products. We build a local large language model pipeline integrated with a text-to-video model, a diffusion model-based system for customised accessory generation, and a GAI-based virtual try-on framework, using open-source tools such as PyTorch, LangChain, HuggingFace, and Diffusers. This study aims to demonstrate how such use cases can be implemented using open source libraries, and not for exhaustive model benchmarking. We identify evaluation metrics for production systems and present deployment estimates for the infrastructure involved. We envision the integrated ecosystem for these use cases forming a coherent flow from interaction to customisation to actual user try-on.
Consumer AIoT devices increasingly embed lightweight language model assistants for smart home monitoring, residential energy management, and consumer energy-awareness applications. However, these models hallucinate sensor values when asked about real-time device state because no temporal synchronization mechanism exists between sensor streams and language model context windows. This paper introduces the Temporal Grounding Pipeline (TGP), a lightweight and secure zero-dependency system that bridges the temporal gap between sensor updates (1–5 Hz) and language model inference (hundreds of milliseconds). TGP comprises four components: a novel in-process circular buffer achieving O(1) operations via Welford’s online algorithm, a dual-criterion staleness detector with F1 = 0.98, a LoRA fine-tuned TinyLLaMA backbone (1.1B parameters, 4-bit quantized), and an optional causal validator. Evaluation across eight consumer AIoT datasets, spanning energy consumption and environmental sensing (commercial buildings, residential homes, household appliances, appliance-level energy, smart-home climate, smart-office occupancy, IoT air quality, and individual consumer electricity), demonstrates 81% average value grounding accuracy compared to 1–9% without grounding, 0.085 ms mean buffer latency (10.4× faster than Redis), and a total memory footprint of 1.09 GB that fits within the envelope of mainstream consumer AIoT hubs and smartphones. All computation remains on-device, ensuring secure, privacy-preserving operation without cloud dependencies.
Detection of harmful gases and their real-time monitoring in closed and open environments find applications such as safety in industrial settings, health monitoring, air quality tracking, etc. This article discusses the issue in the framework of an Internet of Consumer Electronics (IoCE). The model integrates Internet of Things (IoT)-based low-cost self-powered sensor deployment for gas sensing and software-defined network (SDN)-driven control, leading to SDN-IoCE. Reliable transmission of gas sensing data is essential and done through a reconfigurable intelligent surface-aided underlay cognitive radio network. Radio frequency energy harvesting is done for self-powering of sensor nodes. Two deep Q-networks are used, where the first one ensures seamless connectivity on gas sensing data transmission, while the other one performs its time-critical analysis. The proposed SDN-IoT architecture achieves 33-s- and 46-watt-lower values in delay/latency and energy consumption, respectively, over existing work.
The rise of intelligent consumer electronics (ICE), including smart home hubs and wearable devices, requires decentralized mechanisms for secure firmware validation, access control, and tamper-resistant logging. Centralized systems remain vulnerable to spoofed updates, opaque logging practices, and limited scalability. To address these challenges, we present a blockchain-based framework that integrates hash-based firmware verification, smart contract-backed access logging, elliptic curve signatures, and transport layer security-secured transport. Our evaluation, conducted on constrained device profiles (1 vCPU, 512-MB RAM, 10-Mb/s bandwidth), employed four platforms: Hyperledger Fabric (Raft), IOTA (Tangle), Algorand, and Ethereum (proof of authority, PoA). IOTA demonstrated the lowest latency (220 ms) and minimal resource consumption (42-MB RAM, 34% CPU), while Algorand achieved peak throughput (910 tx/s) with 95% consistency. Ethereum-PoA minimized storage requirements (2.8 MB per 1,000 logs), whereas Fabric exhibited higher latency (570 ms) and resource load (74-MB RAM, 51% CPU) but provided strong audit guarantees. Spoofed firmware was rejected in 100% of cases, and unauthorized logs were blocked in over 97.5% of attempts. Privacy was ensured using pseudonymous identifiers and selective logging. Adversarial dynamics, such as packet loss, jamming, and consensus forking were modeled. Despite partial offloading to a gateway node, auditability and verification upheld decentralized trust. These results confirm the feasibility of blockchain-secured ICE and its resilience to both operational and adversarial threats.
With the extensive application of extended reality (XR), intelligent terminals, and the Internet of Vehicles (IoV) to remote areas and disaster regions, immersive consumer electronic applications are driving the demand for global deployment of computing power and low-latency services. To address this challenge, we propose a "cloud center-space edge-air edge-ground terminal" four-layer collaborative computing architecture. We design an intent-driven dynamic resource scheduling mechanism, which directly perceives the quality-of-service requirements of applications through "business intent descriptors." We develop an intelligent computing-aware routing protocol and autonomous offloading decision algorithm to achieve intelligent collaboration of tasks at the terminals, edge nodes, and the cloud. Then, we analyze key enabling technologies such as on-orbit edge computing of satellites and integrated communication-sensing-computing. Simulation results show that the proposed system and mechanisms can significantly enhance the user experience in scenarios such as XR, IoV and smart homes and address challenges such as global coverage and resource scheduling. Finally, we explore solutions to address issues such as dynamic spectrum sharing, communication energy consumption, battery life, protocol standardization, mobility management, and actual deployment. These solutions enable the upgrade of next-generation Internet of consumer electronics.
