Ivana Podnar Žarko

The Internet of Things (IoT) has evolved into the Artificial Intelligence of Things (AIoT), where intelligent data processing complements large-scale connectivity across applications ranging from smart homes to industrial automation. However, its rapid expansion and the increasing adoption of AI have led to growing environmental concerns, particularly increased energy consumption and electronic waste. These issues highlight the importance of Green AIoT practices, which extend Green IoT by combining energy-efficient communication, computing, and intelligent resource management to achieve energy-efficient and sustainable AIoT operation. This paper presents a comprehensive survey of techniques aimed at improving the energy efficiency and sustainability of Green AIoT systems. The focus is placed on networking aspects, particularly machine-to-machine (M2M) communications and wireless sensor networks (WSNs), alongside the roles of computing infrastructures, data centers, and energy-efficient processor architectures. The survey further examines how AI-assisted techniques, including TinyML, edge AI, and intelligent computation offloading, complement traditional Green IoT mechanisms to reduce energy consumption. Key approaches such as low-power communication protocols, energy-efficient data processing, data compression, smart energy management, and energy harvesting are reviewed and compared. Furthermore, the paper summarizes representative state-of-the-art solutions with quantitative insights and discusses open challenges and future research directions toward environmentally sustainable AIoT systems.

Learning-based multi-view 3D reconstruction pipelines, such as transformer-based approaches, enable the accurate reconstruction of 3D scenes from multiple images, but their deployment across the edge–cloud continuum is challenging due to high computational demands and large intermediate data transfers. Effective pipeline scheduling in the continuum must therefore balance latency constraints with the cost of cloud resource usage. In this work, we address cost-aware scheduling under latency constraints for a multi-stage 3D reconstruction pipeline consisting of depth estimation, transformer-based multi-view fusion, and point cloud merging with export to a rendering-ready representation. We implement a service-oriented pipeline where each stage can be executed either on edge or cloud nodes, and we experimentally characterize its performance on representative hardware platforms. The results show a strong imbalance between the computational time and communication latency across platforms, mainly due to large intermediate data. Based on these insights, we propose an online scheduler that dynamically selects stage placements to minimize the cloud cost while satisfying latency constraints. The scheduler incorporates a top-K edge selection mechanism that reduces the decision complexity by jointly considering the network conditions and node utilization. Simulation results parameterized with real-system measurements show that the proposed approach effectively reduces cloud usage while meeting latency constraints, outperforming the baseline strategies based on single-node pipeline execution.

Decentralized trust is a fundamental requirement for blockchain networks where peers lack mutual trust, yet the majority are assumed to be honest for correct network operation. Traditional mechanisms to address this rely on analyzing the blockchain data structure, a process that is resource-intensive for constrained devices. This paper explores Aurora, a novel solution which adds another layer of trust to blockchain networks: It leverages network exploration to achieve trustless and decentralized transaction verification for resource-constrained blockchain clients. Aurora employs a probabilistic algorithm based on hypergeometric distribution to identify sets of peers likely to contain an honest majority. By exploring the network structure, Aurora constructs a set of network peers containing a majority of honest, i.e. trusted, peers with strong correctness guarantees, without relying on a central authority. The probabilistic approach is highly-efficient compared to the deterministic alternative, as demonstrated by analytical analysis and extensive simulations on a realistic Bitcoin-like network. Aurora allows IoT devices to act as ultra-light clients interacting efficiently with the constructed set of peers: An Aurora client checks whether a transaction exists in a valid ledger block by collecting a majority vote from the set without downloading the entire ledger or header chain. Unlike many existing light client algorithms, Aurora seamlessly integrates with existing consensus protocols and offers a novel mechanism for decentralized trust in blockchain networks.

Precision agriculture will continue to heavily rely on data-driven models to enable more intensive crop monitoring and data-driven decisions. The available remote sensing techniques, particularly those based on multispectral Sentinel-2 data, still have major shortcomings due to cloud cover, low temporal resolution, and time lags in data availability. To address these shortcomings, this paper proposes a hybrid approach that combines Sentinel-2 satellite data with real-time data generated by low-cost ground-based single-pixel detectors (SPDs), such as the AS7263. This hybrid approach addresses key shortcomings in existing agricultural monitoring systems and offers a cost-effective, scalable solution for real-time monitoring and prediction of end-of-season yield, moisture, and plant height using simple PLRS models implemented directly in SPDs with an energy-efficient algorithm for deployment on the STM32G030 microcontroller.

The integration of Internet of Things (IoT) technology into agriculture has revolutionized farming practices by using connected devices and sensors to optimize processes and facilitate sustainable execution. Because most IoT devices have limited resources, the vital requirement to efficiently manage data traffic while ensuring data security in agricultural IoT solutions creates several challenges. Therefore, it is important to study the data amount that IoT protocols generate for resource-constrained devices, as it has a direct impact on the device performance and overall usability of the IoT solution. In this paper, we present a comprehensive study that focuses on optimizing data transmission in agricultural IoT solutions with the use of compression algorithms and secure technologies. Through experimentation and analysis, we evaluate different approaches to minimize data traffic while protecting sensitive agricultural data. Our results highlight the effectiveness of compression algorithms, especially Huffman coding, in reducing data size and optimizing resource usage. In addition, the integration of encryption techniques, such as AES, provides the security of the transmitted data without incurring significant overhead. By assessing different communication scenarios, we identify the most efficient approach, a combination of Huffman encoding and AES encryption, to strike a balance between data security and transmission efficiency.

The technology of autonomous driving is currently attracting a great deal of interest in both research and industry. In this paper, we present a deep learning dual-model solution that uses two deep neural networks for combined braking and steering in autonomous vehicles. Steering control is achieved by applying the NVIDIA's PilotNet model to predict the steering wheel angle, while braking control relies on the use of MobileNet SSD. Both models rely on a single front-facing camera for image input. The MobileNet SSD model is suitable for devices with constrained resources, whereas PilotNet struggles to operate efficiently on smaller devices with limited resources. To make it suitable for such devices, we modified the PilotNet model using our own original network design and reduced the number of model parameters and its memory footprint by approximately 60%. The inference latency has also been reduced, making the model more suitable to operate on resource-constrained devices. The modified PilotNet model achieves similar loss and accuracy compared to the original PilotNet model. When evaluated in a simulated environment, both autonomous driving systems, one using the modified PilotNet model and the other using the original PilotNet model for steering, show similar levels of autonomous driving performance.

The paper evaluates the usability of remote satellite-based and proximal ground-based agrometeorological data sources for precision agriculture and crop production in Croatia. The compared agrometeorological datasets stem from the open-access data sources Copernicus CDS and the Agri4Cast portal, and commercial in situ agrometeorological stations (PinovaMeteo) which monitor environmental parameters relevant to the physiological state of crops. The study compares relevant parameters for 10 different locations in Croatia for three consecutive years (2019, 2020, and 2021) to investigate whether model-based data from ERA5-Land and Agri4Cast are well-correlated with ground measurements from independent in situ stations (PinovaMeteo) for specific agrometeorological parameters (air and soil temperature, and precipitation). Our results indicate the following: both the ERA5-Land and Agri4Cast datasets show mostly strong positive correlations with ground observations for air temperature, modest correlations for soil temperature, but modest or even low correlations for precipitation. Analysis of the residuals indicates higher overall residual values, especially in areas with complex topography and near large bodies of water or the sea, and deviations of residuals that may limit the usability of satellite- and model-based data for decision-making in agriculture.

Edge computing is a viable approach to improve service delivery and performance parameters by extending the cloud with resources placed closer to a given service environment. Numerous research papers in the literature have already identified the key benefits of this architectural approach. However, most results are based on simulations performed in closed network environments. This paper aims to analyze the existing implementations of processing environments containing edge resources, taking into account the targeted quality of service (QoS) parameters and the utilized orchestration platforms. Based on this analysis, the most popular edge orchestration platforms are evaluated in terms of their workflow that allows the inclusion of remote devices in the processing environment and their ability to adapt the logic of the scheduling algorithms to improve the targeted QoS attributes. The experimental results compare the performance of the platforms and show the current state of their readiness for edge computing in real network and execution environments. These findings suggest that Kubernetes and its distributions have the potential to provide effective scheduling across the resources on the network’s edge. However, some challenges still have to be addressed to completely adapt these tools for such a dynamic and distributed execution environment as edge computing implies.