The foundation for the Digital Product Passport (DPP) was presented in the recent European Ecodesign for Sustainable Products Regulation (ESPR) in July 2024. DPP systems will provide ICT support to increase the circular economy of products significantly. With the DPP, a significant step forward will be taken by providing comprehensive, trusted, and publicly accessible product information for items or batches regarding digital records. Currently, the DPP building is undergoing a standardization process. Nevertheless, commercial players are preparing DPP systems but mainly with a proprietary implementation, making it difficult for researchers to experiment. In this demo paper, we will show an open-source implementation of a DPP system for managing the lifecycle of second-hand electronic devices such as computers. Our system is oriented towards the use cases of economic actors such as repairers, refurbishers, and the customers of these devices. Our system integrates a blockchain-based registry for the actors to trust the data and operations. The implementation of the demo DPP system is publicly available and can be deployed as a set of microservices.
To strengthen the engineering education at the NORDAKADEMIE, a new collaborative robot laboratory has been developed. This laboratory should provide local access as well as remote access. This paper describes how the remote access is realised using the CrossLab architecture. For this, a Python client is used which implements Server Oriented Architecture Services provided by packages that were developed as part of the CrossLab project. In addition, the challenges and limitations found during implementation are discussed.
This paper introduces PlanARa, a multi-agent system designed for intelligent and adaptive scheduling in an industrial setting. The system leverages agent-based modeling and simulation to dynamically optimize resource allocation, production flows, and task execution while incorporating real-time adaptability to disruptions. The proposed system is designed for seamless integration into real-world manufacturing environments, facilitating dynamic and intelligent scheduling in real-time. Through a simulation-driven approach, the system enables risk-free scenario testing, allowing manufacturers to analyze different scheduling strategies before real-life deployment. Moreover, this system enhances real-time adaptability by dynamically adjusting schedules in response to unforeseen disruptions, thus improving system resilience. The successful implementation of such systems requires careful management of complexity, communication overhead, and integration challenges to ensure scalability and real world applicability. This work showcases how MAS can enhance scheduling efficiency, improve responsiveness to changing demands, and support human-machine collaboration in modern smart manufacturing ecosystems, driving greater flexibility, resilience and efficiency in industrial operations.
Awareness is a major safety factor in human- machine interaction environments, with worker accustomation to robot operation and consequential oblivion of their surroundings potentially having disastrous consequences for either agent or equipment. This is particularly relevant in collaborative scenarios, with workers and robots engaging in close proximity. While consideration for safety is an inherent, albeit forgettable trait of human workers, robots are yet incapable of making such judgements. This translates to collaborative robots typically integrating overt shutdown procedures activated if and when danger is detected for workers, tying with production and other equipment damage issues. Moreover, human focus is both intrinsically tied with safety judgement and manifestable in variable bio-signalling, of which some can be captured via brain-computer interface technologies. This short paper advocates for the evaluation of worker bio-signalling to avert danger, by instead adapting robot operation. It presents the monitoring of human attention as a by-product of single channel electroencephalography waves, to modulate robot functioning to complete its task, track its human colleague while they do their part, or halt. Implications and limitations are pondered for usage in real-life industrial scenarios.
The Remote FPGA Motion Lab is a remotely accessible platform designed for hands-on experimentation in digital systems design and embedded control. It is based on the FPGA Motion Lab, a physical experimental setup for FPGAbased system development. Through LabsLand LabDiscovery Engine, it offers a scalable Remote Laboratory Management System (RLMS), managing user authentication, session queuing, and concurrent experiment instances. The interface allows cloud-based VHDL synthesis or bitstream upload, providing full control over motors, sensors, and GPIOs. The Remote FPGA Motion Lab will be tested in the 2024-2025 academic year for the Programmable Logic course at University of Deusto.
Servomechanisms present an excellent opportunity for introducing topics such as dynamical systems modelling, system identification, control theory and the PID controller. This paper describes an attempt at conciliating theoretical and simulated system's responses, which are typically developed in the continuous-time domain, with results from experiments that use a digital implementation of the PID controller. An example application of proportional position control of a brushed DC motor illustrates how the direct comparison between simulation and experiment results can be used to augment the learning experience in laboratory classes.
In the last decade the effects of climate change, increasing levels of traffic and the need to move to a circular and sustainable economic strategy have led to a more complex and interdisciplinary design and management strategy of transportation infrastructure. Because of that, many international, national, and regional agencies are funding and promoting interdisciplinary based projects. This development underlines the need to integrate workplace dynamics into learning, ensuring that education remains aligned with the realities and demands of the professional environment. In recent years, European Union has been shedding light on advancing asphalt pavement engineering through a series of research projects under the umbrella of Horizon 2020 and the European Regional Development Fund. In light of these developments, there is a growing need for advanced master's programs that introduce students - the future workforce, to industry-applied processes while maintaining strong connections to the world of work through close cooperation between the education sector and industry. This paper aims to investigate whether asphalt pavement-based master's programs exist and the need for such programs. Particularly, this study aims to review the existing joint and cross-country master's programs offering specific, transversal, and multidisciplinary courses for pavement engineering. The review is based on a global search of university programs and a detailed examination of their curricula, by performing an online search assisted by the generative AI tool ChatGPT. Finally it proposes the need for a joint master's program relevant to pavement engineering with a focus on technological advancements and sustainability.
Social robots are increasingly being deployed in indoor environments such as hospitals or malls, to assist with navigation and give information related to indoor spaces. Therefore, these social robots need to understand the human questions using a natural language understanding and generation, making them suitable for enhancing social robots' interaction capabilities, also, the need to understand the indoor environments that need to be mapped. Large language models can be useful for understanding the semantics associated with questions asked by users and the answers generated by the social robot. This paper explores the integration of LLMs into social robots for indoor location navigation. We describe the architecture of our LLM-powered social robot, its capabilities, and challenges related to real-time processing, user adaptation, and accuracy. A pilot study evaluates the robot's efficiency in guiding 20 users and responding to queries in a indoor environment.
Lifestyle behaviours play a crucial role in shaping an individual's overall well-being and health. In an increasingly sedentary society, it is important to promote active and healthy routines. This study contributes by investigating how environmental and behavioural factors influence physical activity and mental well-being. A public dataset available on Kaggle was used, combining biosignals, GPS coordinates, accelerometer data and self-reported information. Although the dataset was limited, containing data from only two users, it allowed the identification of key contextual features that influence lifestyle classification. These included location, companionship, type of movement, and the presence of green spaces or streets and avenues, all of which were found to be determinants of an active and positive lifestyle. Based on these features, three machine learning models - Random Forest, Logistic Regression and XGBoost - were developed and achieved classification accuracies of over 75% in identifying lifestyle type. Despite these promising results, further research is needed to incorporate more comprehensive data and additional factors to improve model performance and generalisability.
Robotic automation is an integral part of modern industry. It is thus imperative to prepare students of engineering and computer science to handle industrial robots. However, as stopping a single machine can impact the productivity of an entire facility, training on live machinery might not always be feasible. Likewise, setting up enough laboratory space and equipment for students to use during their studies requires much available space and funding and thus also might not always be feasible. This paper presents a prototype of a digital twin for a collaborative robot arm. The digital twin makes use of augmented reality for visualisation and connects to either a physical robot or a simulated robot controller inside a virtual machine. Several aspects of the implementation as well as limitations and lessons learned are described.
The rapid progression of autonomous driving and advanced driver-assistance systems has emphasized the need for robust and efficient object detection within vehicle environments. This paper conducts a comparative analysis of state-of-the-art object detection models for in-vehicle environments using public datasets, guided by the CRISP-DM (Cross-Industry Standard Process for Data Mining) methodology. The study includes several YOLO models (YOLOv5, YOLOv7, YOLOv8, and YOLOv10), a transformer-based model (RT-DETR), and a CNN-based model incorporating self-attention mechanisms (YOLO-NAS). The evaluation relies on key metrics such as mean Average Precision (mAP) and inference time to determine each model's effectiveness. The experimental results indicate that while our approach to vehicle interior objects detection for forgotten items shows promise, the overall performance fell short of expectations. The YOLOv7 x variant achieved a recall of 0.632 and an mAP50 of 0.351, offering better precision than YOLOv5 in some cases. However, its 10.3-millisecond GPU inference time underscores the challenge of balancing accuracy and computational efficiency. The findings emphasize the impact of dataset quality and class balance in achieving robust and reliable object detection, highlighting the need for tailored datasets to enhance model reliability in real-world automotive applications. This research contributes to the broader vision of green and smart cities by advancing technologies that improve vehicle safety and support sustainable urban mobility systems.
Online and automated psychological interventions gained increasing attention due to their accessibility and cost-effectiveness compared to traditional therapeutic approaches. Among these, breathing techniques that can induce cardiac coherence, which have shown benefits in regulating autonomic functions and enhancing interoceptive awareness in traditional settings, might be valuable approaches to be used in web-based interventions. This pilot study aimed to assess the feasibility and potential psychophysiological effects of an automated breathing exercise, delivered with an audio clip, designed to increase cardiac coherence. Forty-one healthy volunteers completed an online intervention that included pre and post measures of relaxation, emotional state, and body awareness using self-report scales. Results showed significant improvements in overall relaxation, emotional experience and interoceptive awareness. These findings support the feasibility of digital cardiac coherence interventions and highlight their potential benefits for psychophysiological well-being.
Ondroid is a cloud-based remote robotics laboratory designed to enhance hands-on learning in robotics, ICT, and automation. The system enables students to program and interact with real robotics hardware remotely, with assignments automat-ically verified through computer vision and sensor data analysis. It is structured to allow simultaneous access to a shared hardware environment for multiple students, providing cost-effective and scalable remote education. This paper details the architecture of this system, its technical implementation, and its application in education. The conference demonstration showcases real-time programming of an ESP32-based mobile robot using C++ and a Dobot MG400 robotic arm using Lua.
Technological advancements have transformed interactions, affecting socio-cognitive processes and overall communication experiences. This study investigates the impact of communication modalities - face-to-face (FTF) versus remote work (RW) - on collaborative performance and interpersonal closeness. 48 participants were paired into dyads and engaged in collaborative tasks under both FTF and RW conditions. Participants performed two cooperative tasks, namely, the Lost at Sea/Desert Survival task, to measure cooperation and shared decision-making abilities, and an ecological work problem simulation in which they had to allocate a budget for employee training, to assess practical collaboration and problem-solving skills. Participants also completed standardized questionnaires assessing media richness, interpersonal closeness, and perceived performance. Results indicate that FTF communication is perceived as richer and more socially engaging compared to RW, facilitating greater social connection between participants. However, no significant differences emerged in perceived performance or collaboration scores. These results suggest that FTF interaction is perceived as richer as RW, facilitating stronger interpersonal connections and greater social engagement. Future research could further investigate the role of individual and contextual variables to better understand the conditions that optimize collaboration in FTF and RW, through the use of more complex measures and by involving larger groups.
Stroke rehabilitation is essential for restoring motor function and improving the quality of life for survivors. Robotic-assisted therapy has the ability to provide accurate, repeatable, and adjustable exercises for recovery. This study investigates the feasibility of using a UR5 collaborative robot for stroke rehabilitation, integrating force control and position/velocity control to facilitate structured rehabilitation exercises. The system is designed to provide variable assistance, adjusted based on the patient's ability, ensuring a personalized rehabilitation experience. The rehabilitation setup includes force-controlled tracing exercises on a whiteboard for visual feedback and position/velocity-controlled reaching tasks that mimic real-life movements. Preliminary patient testing showed effective motor activity, task-based learning, and flexibility. The system's ability to provide real-time performance feedback and configurable exercises highlights its potential in improving rehabilitation outcomes. However, challenges such as optimizing resistance levels and refining task difficulty progression remain. Future research will focus on real-time adaptive control strategies, enhanced feedback mechanisms, and gamification to increase patient motivation.
The objective of this paper is to evaluate the impact of Artificial Intelligence techniques on the prediction and improvement of teaching performance in Computer Science, in order to provide a basis for the design of educational policies. Approaches such as Bayesian classifiers, support vector machines, and decision trees are analyzed. The research uses a database of 2,335 records and 19 attributes, corresponding to 109 groups, obtained from surveys on teaching performance at a university in Colombia. The results show that decision trees are the most effective models, with an accuracy of over 86%, when cross-validation by folds (50%-50%) is used through the Weka platform. Additionally, the main independent variables identified are: effort made to ensure everyone learns, adequate preparation of classes, pleasantness of work style, and encouragement of the topics covered. Finally, this study achieves a 55% reduction in independent variables and examines the implications of these techniques in the design of educational policies
Sustainable urban mobility, promoted by shared vehicles, has gained attention as a solution to reduce carbon emissions and facilitate transportation in cities. However, man-aging these vehicles presents challenges, such as internal damage caused by vandalism or improper use. This study evaluates the effectiveness of different Artificial Intelligence (AI) approaches for automatic interior damage detection in shared vehicles, considering their specific characteristics, such as compact and urban-oriented designs. The research also investigates whether the public dataset used, consisting of images of vehicle interiors with and without damage, is suitable for this scenario, given that shared vehicles may differ from conventional cars. Convolutional Neural Network (CNN) approaches, such as YOLOv5, YOLOv7, YOLOv8, YOLOv10 and RTMDet, as well as transformer-based models like RT-DETR and Dino, and hybrid approaches incorporating attention mechanisms, such as YOLO-NAS, were analyzed. The key question is whether damage detection in shared vehicles can generalize to highly variable images in this dataset. The diversity of vehicle interiors, including differences in materials, lighting conditions, and structural configurations, presents a significant challenge for AI. Ensuring accurate damage detection despite this variability is crucial for practical applications. Therefore, it is essential to assess not only model accuracy but also their ability to generalize to new vehicle configurations not represented in the dataset. Furthermore, this study provides insights into the necessary next steps for developing a robust and effective damage detection service, contributing to better management and maintenance of shared vehicle fleets.
Supported by different activities and examples inte-grating augmented reality (AR) technology and associated results in engineering education and training, this study pretends to highlight the benefits and potential advantages of this approach. By its use in different teaching and learning dimensions, AR can positively influence informal education, serve as a beneficial tool with promising perspectives, and be used in degree and at the graduation projects to provide insights into labor environments. This work emphasizes the relevance of AR in shaping perceptions for future professionals, enhancing their learning experi-ences and preparing them for their future in modern professional environments. It includes various examples illustrating the perspectives tested across different scopes, based on the teaching activities of the last three authors and the first author's experience as a PhD student. These experiences were tested at different levels of mechanical engineering education: pre-university programs, undergraduate, master's, and post-graduate levels.
In critical areas with significant user impact, the effective application of machine learning models to support decision making faces the important challenge of lack of trust. Despite their impressive performance, machine learning models are often perceived as “black boxes,“ making their adoption difficult. This is the case of health care area, where the lack of trust is undoubtedly a barrier to the application of those models in daily clinical practice. While metrics like accuracy, sensitivity, specificity are valuable to assess the global performance of a model, they provide little insight at the individual instance level (single patient). This highlights the importance of pointwise reliability assessment, evaluating whether a model can be reliable to classify a specific instance. This work aims to develop a systematic method for comparing a machine learning model pointwise reliability provided by different methods. Besides error rate quantification, this framework comprises t-Distributed Stochastic Neighbor Embedding visualizations. Thus, different data driven methods for assessing pointwise reliability are compared. These methods address the problem based on two different perspectives: i) density principle; ii) local fit principle. The validation was performed in the context of cardiovascular secondary prevention supported by a real patient dataset (N=1544). The findings show that approaches integrating both density and local fit principles tend to surpass single principle methods, associating lower error rates to more reliable predictions.
Machine learning is nowadays used in tiny micro-controller boards, sometimes called TinyML. At the same time, LoRa Internet of Things (IoT) communication technology can now provide LoRa mesh network topologies that interconnect remote IoT devices. Both tendencies, however, have not converged yet to combine the potential of both. In this paper, we argue for a solution consisting of a two-board design as an architecture for interconnected ML-enabled IoT nodes in a LoRa mesh network. A prototype of the architecture using the TTGO LoRa32 board as a network device and the Arduino Portenta H7 as an application device is implemented as a case study to show the feasibility. We analyze a federated learning protocol design to illustrate how distributed applications can interact within the proposed two-board node architecture. We position the architecture with regard to several performance aspects and observe that the design delivers a separation of concerns by one board hosting the ML-application and the other providing an advanced network service to the application.
