Use cases

Use case description – Data Needs: The development of AI-based intelligent control systems in manufacturing, with functionalities such as zero-waste manufacturing or predictive maintenance, requires collecting process and operations data from the assets at operation stage for the AI models to be trained and tested. These applications require quality labelled data, moving the focus to the improvement and enrichment of datasets employed to train the algorithms rather than to the algorithms themselves. In this concept, three key aspects need to be carefully considered:

(a) the quality of heterogenous data, (b) the continuous evaluation of heterogenous data, as contexts in manufacturing scenarios are never static and data drift, which need to be detected and adapted, (c) the integration of expert human knowledge (Human-in-the-loop), since domain experts can determine whether a dataset accurately represents the problem at hand. In PLIADES, during operation of the metal forming, machine data from machine control systems i.e., sensors, PLCs, cameras, etc. either raw or processed (e.g., KPI’s, AI services for quality and maintenance results) will be transferred to the industry data space and deployed by the means of the proposed framework.

PLIADES framework: Through the context and cultural-aware data analysis methods, such as human-in-the- loop approaches, to AI development, such as active learning, interactive learning or machine teaching, expert human domain knowledge will be efficiently and transparently integrated into machine data, e.g., to ensure accurate labelling. In that sense UC real on-the-field information will be inserted in our data storage system. This data will be then made available, in an appropriate way, to industrial end-users or environmental management entities via our AI-based data brokering schemes, based on DataOps approach for data lifecycle management workflows design. The data users will then use our data integration framework so as to find these data, further elaborate and use them appropriately, in order to: improve the ML/DL models of their own machine systems or develop new environmental assessment and optimization systems, while the sovereignty of the data will be appropriately maintained at all stages of the full data life cycle.

Use case description – Data Needs:Assistive service robots, either for rehabilitation and patient support purposes in hospitals or for eldercare support in nursery or domestic settings, need data, collected during their actual operation, for their HRI capabilities to be trained and improved over time. The individual specificities of each human make each HRI session rather unique while the recent advances in DL methods highlight the need for large volumes of data to train AI systems that operate with highly advanced capabilities of personalization and effectiveness during the interaction. In PLIADES, the collection of massive data that can be used to train human activity, behavior monitoring and efficient HRI methods for (a) patient support and rehabilitation robots – BOR, and (b) social assistive service robots – CERTH, will be explored. In particular, the data that are created from the robot sensors during the operation of the rehabilitation or the social assistive service robots, i.e. from the robot’s camera, lidar, microphone and further sensors, either raw or processed, w.r.t. human body tracking, activity recognition and human speech, will be transferred in the healthcare data space deployed by the means of the proposed framework. 

PLIADES framework: The data will be catered for human factors and cultural issues to become properly homogenized when inserted in our distributed data storage system through the context and cultural-aware data analysis methods for human tracking, as well as through human data processor intervention when necessary, e.g., to ensure precise data labelling. These data will be then made available via our AI-based data brokering schemes, in an appropriate way, to relevant robotic technology providers, be those the same as the data owners, or different ones. The data users will then use our data integration framework to find these data, further elaborate and use them appropriately, to improve the ML/DL models of their own robotic systems, while the sovereignty of the data is appropriately maintained at all stages of the full data life cycle.

Use case description – Data Needs: The research and development of personalized medicine approaches requires the collection of a vast amount of personal healthcare data in a safe, anonymized and efficient manner. In this scope, MONDRAGON Corporation (with MU-EPS as cooperative university) together with CICBiogune are currently developing a data platform capable of storing, managing and visualizing large volumes of data from over 11.000 participants. The collected data refer to biochemical, metabolomic, proteomic, genomic, lipidomic, lifestyle (nutritional and activity) and sociodemographic information while several API services have been developed for data consumption, research questions definition, data processing and visualization. In PLIADES, the developed platform will be integrated in the proposed framework through a specific connector that attends to the IDSA requirements.

PLIADES framework: The data will be properly homogenized when inserted in our distributed data storage system through the context and cultural-aware data analysis methods for personalized medicine as well as through human data processor intervention when necessary, e.g., to ensure precise data labelling. This data will be then made available in the healthcare data space, via our AI-based data brokering schemes, in an appropriate way, to relevant healthcare end users such as health science researchers, software service developers, health prevention services providers and drug companies. Finally, the data users will be able to use our data integration framework to discover relevant data, further elaborate and use them appropriately to enhance their AI services that improve the diagnostic and prognostic clinical prediction models in the healthcare sector.

Introduction – Data Needs: From the currently more mainstream levels of automation to the much-anticipated higher levels, the automotive industry currently relies on highly advanced ML and DL networks to train the smart functions required for efficient autonomous vehicle perception and driving/control methods to be developed. To achieve it, more and more data from the vehicle’s sensors, lidars, cameras, as well as from further sensors including roadside infrastructure, concerning the vehicle’s state and a vast amount of proprioceptive information is required while it becomes clear that the core step at this point to boost vehicles’ autonomy is the deployment of further advanced technological means that enable the collection, treatment, use and re-use for (re-)training of ML and DL models, of data taken from autonomous vehicles’ sensors in diverse street settings, from clearly structured through to unstructured environments. In PLIADES, during the operation of the autonomous vehicles of Taltech and CERTH, the data that are created from the vehicle sensors, either raw or processed, concerning street elements and humans’ detection, as well as vehicle state and ADAS system decisions, are transferred in the mobility data space deployed by the means of the proposed framework. Vicomtech and AVL using their vehicle will prepare an edge-cloud processing pipeline for raw data management from sensors and functions to be shared in the Mobility Dataspace synchronizing development cycles of car makers and ADAS function providers. CVUT will then follow with methods for enriching vehicle data by roadside infrastructure data and data from floating vehicles serving as similar type of sensors. Real-time access to anonymized vehicle and infrastructure data will allow ZERO to improve traffic safety, reduce congestion, and increase time and energy efficiency of transport. Additionally, it will provide an open platform for users to improve their driving behavior, acquire scientific data, and optionally share V2X data through an aftermarket device or smartphone.

PLIADES framework: The data will become properly homogenized, abstracted and semantically annotated, when inserted in our distributed data storage system through the context and cultural-aware data analysis methods for scene, vehicle state and decisions analysis as well as through human data processor intervention when necessary, e.g., to ensure precise data labelling. These data will be then made available via our AI-based data brokering schemes, in an appropriate way, to relevant ADAS technology providers, be those the same as the data owners, or different ones (e.g., AVL, CERTH, VICOM). The data users, will then use our data integration framework to find these data, further elaborate and use them appropriately, to improve the ML/DL models of their own ADAS/AD systems, improve safety and reduce congestion, while the sovereignty of the data will be appropriately maintained at all stages of the full data life cycle.

Use case description – Data Needs: For automotive engineering service providers, the use of data intelligence is about bridging the gap between the digital and the engineering worlds to help OEM-s and Tier-1-s to leverage the use of data from vehicles as well as the eco-systems around the vehicle. In this testbed, we address the issue of prognosis and health management in EVs. To cope with the 5 V’s and multi-physics (mechanics, thermal, electric, etc.) nature of (mostly) time-series data, we need to extract vast amount of data referring to driving, charging events, etc. and calculate connected aggregates in a dedicated analytics engine. Events and aggregates need to be stored for sorting and re-use to enable fast and interactive analysis. In PLIADES, the integration of Dev/ML-Ops approaches and development of new EV data processing and management tools for predictive actions in EV will be explored. The predictive capabilities will be exercised when it comes to the prediction of the health status of components in the EV powertrains, continuous prediction of the state of health (SoH) of batteries and continuous prediction of the integrity on mechanical components.

PLIADES framework: The data will be catered for human factors (considering the different roles and background of the data consumers) to become properly homogenized and ready to be consumed by any professional through the context -aware data analysis methods for human interpretation, as well as through human data processor intervention when necessary, e.g., to ensure precise data labelling. Moreover, via our AI-based data tools, we aim at creating a Machine Learning data flow as a methodology to manage the lifecycle of data inside the platform. Traceability and data provenance shall be supported by graphical databases that register metadata about source, modification, confidence, value or quality of data while building a transparent, fair, and trustworthy AI framework that will address both safety and explainability aspects without compromising on quality assurance. The data users can then use our data integration framework to find these data, further elaborate and use them appropriately, while the sovereignty of the data is appropriately maintained at all stages of the full data life cycle.

Use case description -Data Needs: Professional service robots need to be endorsed with further advanced capabilities for smooth and efficient HRI to promote their adoption in real healthcare and industrial environments. Although deep-learning -based approaches gain more and more ground in this scope, they require vast amounts of relevant data to be collected and become available for re-use to achieve effective training. Moreover, these data refer to human commands from different modalities (e.g. speech, gestures) that have to be translated in specific robot plans and actions. Nevertheless, they might include different levels of abstraction or be affected by significant human factors w.r.t. individual and cultural specificities. In PLIADES, HRI data of three different scenarios will be considered: (a) a robot that performs patient support and rehabilitation in a healthcare setting – BoR, (b) robots that performs nursing personnel support – performing disinfection and monitoring of patients in a healthcare setting – BoR, CERTH and (c) a robot that performs telepresence-based inspection in a manufacturing setting. The data that are created by the robot sensors in the aforementioned scenarios w.r.t. the commands that the robots receive, will be transferred in the healthcare and manufacturing/industrial data spaces deployed by the means of the proposed framework.

PLIADES framework: The data will be catered for human factors and cultural issues to become properly homogenized when inserted in our distributed data storage system through the context and cultural-aware data analysis methods for human inputs monitoring (speech, gestures, etc.) as well as through human data processor intervention when necessary, e.g., to ensure precise data labelling. These data are then made available, via our AI- based data brokering schemes, in an appropriate way, to relevant robotic technology providers, be those the same as the data owners, or different ones, both in the healthcare and the industrial data spaces, with further possibilities to extend the scope, by involving e.g., the agricultural data space as well. Finally, the data users can use our data integration framework to discover relevant data, further elaborate and use them appropriately, to improve the ML/DL models of their own robotic systems, while the sovereignty of the data is appropriately maintained at all stages of the full data life cycle.