Search

The learning of new knowledge and skills often requires previous knowledge, which can lead to some frustration if a teacher does not know a learner's exact knowledge and skills and therefore confronts them with exercises that are too difficult to solve. We present a solution to address this issue when teaching techniques and skills in the domain of table tennis, based on the concrete needs of trainers that we have investigated in a survey. We present a conceptual model for the representation of knowledge graphs as well as the level at which individual players already master parts of this knowledge graph. Our fine-grained model enables the automatic suggestion of optimal exercises in a player's so-called zone of proximal development, and our domain-specific application allows table tennis trainers to schedule their training sessions and exercises based on this rich information. In an initial evaluation of the resulting solution for personalised learning environments, we received positive and promising feedbackfrom trainers. We are currently investigating how our approach and conceptual model can be generalised to some more traditional educational settings and how the personalised learning environment might be further improved based on the expressive concepts of the presented model.

In this work, we introduce a novel pipeline that enables the generation of multiple-choice questions and exercises from as little as a topic keyword. Hence, providing users the possibility to start with a study objective in mind and then automatically generate personalized learning material. The main contributions of this project are a scraper that can extract relevant information from websites, a novel distractor generation method that can make use of context and a technique to automatically combine text and questions into interactive exercises. Our novel distractor generation method was tested in a human survey which showed that the distractor generation quality is comparable to hand crafted distractors. The pipeline is built into a web application that lets users refine the results for each step, openly accessible at https://adaptive-teaching.com., CSEDU: Proceedings of the 14th International Conference on Computer Supported Education, ISBN:978-989-758-562-3

The intake of computer science faculty has rapidly increased with simultaneous reductions to course personnel. Presently, the economy is recovering slightly, and students are entering the working life already during their studies. These reasons have fortified demands for flexibility to keep the target graduation time the same as before, even shorten it. Required flexibility is created by increasing distance learning and MOOCs, which challenges students’ self-regulation skills. Teaching methods and systems need to evolve to support students’ progress. At the curriculum design level, such learning analytics tools have already been taken into use. This position paper outlines a next-generation, course-scope analytics tool that utilises data from both the learning management system and Gitlab, which works here as a channel of student submissions. Gitlab provides GitOps, and GitOps will be enhanced with machine learning, thereby transforming as MLOps. MLOps that performs learning analytics, is called here LAOps. For analysis, data is copied to the cloud, and for that, it must be properly protected, after which models are trained and analyses performed. The results are provided to both teachers and students and utilised for personalisation and differentiation of exercises based on students’ skill level. publishedVersion

Interest in and demand for micro-credentials in higher education institutions is on the rise. Although the concept of micro-credentials is still evolving, they can be seen as short learning opportunities that are accompanied by digital credentials that capture the proofs of the learning. These digital proofs of learning range from skills and competences acquired to information whether such skills were acquired via formal or non-formal learning activities. Micro-credential platforms are used for multiple purposes including issuing, viewing, and storing the digital credentials. Despite the growth in the number of micro-credential platforms in the recent years, literature is limited on the features offered by the platforms and how they are helpful for higher education institutions and learners. To address this gap in research, we employed a qualitative approach by semi-structured interviews and group discussions with platform providers and education experts. Our findings resulted in 38 features that can help higher education institutions, learners, and providers understand what kind of features are emphasized in micro-credential platforms and how they can be helpful for different use purposes. As practical implications, the findings of this study can help higher education institutions in considering adoption and usage of micro-credential platform. publishedVersion

Although logic is considered central to mathematics and computer science, there is evidence that teaching logic has not been a great success. We identify three issues where what is typically taught conflicts with what is needed by those who are supposed to apply logic. First, what is taught about the notion of implication often disagrees with human intuition. We argue that in some cases human intuition is wrong, and in some others teaching is to blame. Second, the formal concepts of logical consequence, logical equivalence and tautology are not the similar concepts that everyday mathematicians and computer scientists need. The difference is small enough to go unnoticed but big enough to cause confusion. Third, how to deal with undefined operations such as division by zero is left informal and perhaps fuzzy. These problems also harm development of computer tools for education. We present suggestions about how to address them in teaching. peerReviewed

The recent generations of massive spectroscopic surveys aim at the ray collection from a multitude of cosmological targets in the course of observations. For this purpose, astrobots are used to change the configuration of optical fibers from one observation to another in relatively short periods of time instead of tedious manual replacements. The dense formations of astrobots on focal planes enhance the number of the potential targets to be observed. However, the safe coordination of astrobots swarms is challenging. The more astrobots are coordinated, the more data are sent to a spectrograph, thereby the higher the resolution of a resulted survey will be. However, traditional collision-avoidance coordination strategies often give rise to the partial convergence of astrobots swarms. Thus, this thesis focuses on the solutions to the complete safe coordination of astrobots, particularly in the case of Sloan Digital Sky Survey V. We increase coordination convergence rates not only by directly improving the state-of-the-art coordination solution but also by optimizing target-to-astrobot assignments. Namely, we propose an optimal assignment scheme which minimizes both the likelihood of collisions between astrobots and the effort demanded to preform coordination in terms of the required time to perform coordination. We also propose a cooperative coordination method in which, given particular settings of astrobots and/or targets, each astrobot stops at its goal point when its other neighboring peers have already reached theirs, as well. So, we derive a localized completeness condition that, if sufficed, generates the trajectories which completely coordinate an astrobots swarm in a guaranteed manner. We also propose a logic-based formally-verifiable supervisory coordination technique whose behavior is always safe and complete without any need to simulation-based validations. Finally, we employ machine learning tools to train models to predict the feasibility of complete coordination only according to initial and final configurations of astrobots and their targets' projected locations on their focal plane. These models contribute to the identification of those target-astrobot pairings which do not fulfill the completeness condition. So, one may re-plan such ill-posed assignments before getting involved in potentially-pointless completeness checking simulations. Our simulated and experimental tests manifest the efficiency of the developed methods quoted above.

Driving is a very challenging task to automatize despite how naturally and efficiently it may come to experienced human drivers. The complexity stems from the need to (i) understand the surrounding context and forecast how it is likely to evolve, (ii) plan motions so that maneuvers can be performed with a certain level of anticipation despite the uncertainty of the future traffic state, and (iii) act on the throttle and the steering wheel to execute the planned motions accurately. These tasks match the major research topics concerning autonomous driving, namely perception and prediction, motion planning, and control. In this thesis, we study challenges related to motion planning and decision-making for connected automated vehicles (CAVs) in mixed traffic. That is for CAVs that coexist with human drivers, other CAVs, and unconnected automated vehicles (AVs). Even though we intend to formulate the proposed methods so that they are context agnostic, their assessment is carried out in roundabout scenarios. Roundabouts are ideal testing scenarios due to the complexity of the traffic interaction and overall traffic dynamics, the impact that uncertainty has on the coordination performance, as well as the strong influence that dynamic occlusions of the surroundings caused by nearby vehicles have on the decision-making process. We propose a novel approach concerning how an AV's surrounding space is represented and described, which brings benefits to the motion planning module. Unlike the classical planning approach based on object detection and avoidance, we study an alternative strategy based on free space identification and exploitation, which is shown to be a suitable mechanism to account for occlusions and other perception uncertainties. Our planning solutions are model-based and--inspired on the way human drivers seem to make decisions--aim to make safe yet efficient decisions without the need to explicitly explore all possible trajectories that can be followed. Instead, we propose a low-dimensional driving maneuver representation that enables us to characterize the solution-space of the decision-making problem at a high-level. In particular, a novel planning framework is presented in this thesis to address four significant planning aspects. Firstly, a reactive gap-acceptance behavior is formulated, which represents an appropriate baseline behavior despite its simplicity. Afterward, we investigate a decision-making approach for CAVs in fully connected scenarios, whereby CAVs would consider the impact of their decisions on the overall traffic before executing them. Then, we address the challenge of making AVs cooperate with other unconnected vehicles through a so-called implicitly cooperative mechanism. Furthermore, we present a predictive-reactive planning strategy where the challenge of planning motions taking into account longer traffic predictions, and the possibility of them being wrong is tackled. Finally, the suitability of some of the proposed theoretical results is assessed in a more realistic setup, where the methods are applied to real data provided by our industrial partners. This dissertation provides new ideas and methods to address the complexity of motion planning in mixed traffic. Specifically, we tackle the problem through a versatile motion planning framework and a set of pragmatic model-based decision-making strategies, paving the way towards feasible, efficient, and more reliable solutions.

Social media (SM) platforms have demonstrated their ability to facilitate knowledge sharing on the global scale. They are increasingly often employed in educational and humanitarian domains where, despite their general benefits, they expose challenges peculiar to these domains. Specifically, the research context of this thesis is directed by my participation in the Go-Lab European project and my collaboration with Médecins Sans Frontières (MSF) where SM platforms were used extensively. In this thesis, we address four challenges regarding analytics, privacy, discovery, and delivery, aiming to answer corresponding four research questions. How to provide user-oriented analytics in knowledge sharing systems to support awareness and reflection? What privacy management interfaces and mechanisms are suitable for knowledge analytics and learning analytics? How to enable discovery of knowledge relevant to user interests? How to facilitate knowledge delivery into settings where Internet connectivity is limited or absent? Henceforward, we provide an overview of our results. Analytics. To enable awareness and reflection for an SM platform users, we propose the embedded contextual analytics model where the analytics is embedded into the interaction context and presents information relevant to that particular context. Also, we propose two general architectures materializing this model respectfully based on real-time analytical applications and a scalable analytic back-end. Using these architectures, we provided analytics services to the SM platform users. We conducted an evaluation with the users demonstrating that embedded contextual analytics was useful to support their awareness and reflection. Privacy. To address the privacy concerns associated with the recording, storage, and analysis of user interaction traces, we propose a novel agent-based privacy management model. Our proposal uses a metaphor of physical presence of a tracking agent in an interaction context making the platform user aware of the tracking and allows to manage the tracking policy in a way similar to the physical world. We have implemented the proposed privacy interface in an SM platform and obtained positive evaluation results with the users. Discovery. Due to a large number of content items stored in SM platforms, it can be challenging for the users to find relevant knowledge. Addressing this challenge, we propose an interactive recommender system based on user interests enabling discovery of relevant content and people. We have implemented the proposed recommender in an SM platform and conducted two evaluations with platform users. The evaluations demonstrated the ability of the approach to identify relevant user interests and to recommend relevant content. Delivery. At the moment of writing in 2016, near half of the world's population still does not have reliable Internet access. Often, the places where humanitarian action is needed have limited Internet connection. We propose a novel knowledge delivery model that relies on a peer-to-peer middleware and uses low-cost computers for local knowledge replication. We have developed a system implementing the model and evaluated it during eight deployments in MSF missions. The evaluation demonstrated its knowledge delivery abilities and its usefulness for the field staff.

Learning Analytics is about collecting traces that learners leave behind and using those traces to improve learning. Dashboard applications can visualize these traces to present learners and teachers with useful information. The work in this paper is based on traces from an inquiry-based learning (IBL) environment, where learners create hypotheses, discuss findings and collect data in the field using mobile devices. We present a work-in-progress that enables teachers and learners to gather around an interactive tabletop to explore the abundance of learning traces an IBL environment generates, and help collaboratively make sense of them, so as to facilitate insights. ispartof: pages:32-35 ispartof: Proceedings of the First International Workshop on Visual Aspects of Learning Analytics co-located with 5th International Learning Analytics and Knowledge Conference (LAK 2015) vol:1518 pages:32-35 ispartof: Visual Aspects of Learning Analytics location:Poughkeepsie, NY, USA date:17 Mar - 17 Mar 2015 status: published

In this paper, we briefly describe the goal and activities of the LAK15 workshop on Visual Aspects of Learning analytics. ispartof: pages:394-395 ispartof: Proceedings of LAK15: Fifth International Conference on Learning Analytics And Knowledge vol:16-20-March-2015 pages:394-395 ispartof: International Conference on Learning Analytics And Knowledge location:Poughkeepsie, NY, USA date:16 Mar - 20 Mar 2015 status: published

The coordination of multi-robot systems is becoming one of the most important areas of research in robotics, mostly because it is required by numerous complex applications. These applications range from intelligent transportation systems, search and rescue robots, and medical robots, to cosmology and astrophysics. The coordination of multi-robot systems is based upon cooperation. The actions performed by each robot take into account the actions executed by the others in such a way that the whole system can operate coherently and efficiently. Regardless of the application, coordination is the key to the successful design and implementation of multi-robot systems. The number of robots involved in the aforementioned applications is increasing along with advances in miniaturization and automation. Consequently, a large number of robots need to share a workspace. This crowded workspace introduces new challenges into the coordination problem by increasing the risk of collision. To take into account communication constraints and sensor ranges, robots rely on local information. Therefore, efficient but simple coordination algorithms are required. This thesis investigates decentralized approaches based on navigation functions for the coordination of multi-robot systems in crowded workspaces. Decentralization allows robots to rely on local information, guarantees scalability and enables real-time deployment. Navigation functions are a special category of potential functions. Their negated gradient vector-field is attractive towards the goal and repulsive with respect to fixed or moving obstacles to avoid collision. In the first part of the thesis, we present the multi-robot coordination problem using navigation functions in a game-theory based framework. We propose a motion model along with a control law that leads the robots to a Nash equilibrium. The existence of the Nash equilibrium enables navigation functions to be exploited for studying, building, and running coordination frameworks for multi-robot systems. In the second part, we address the coordination of autonomous vehicles at intersections. A novel decentralized navigation function is proposed. It guarantees collision-free crossing of autonomous vehicles modeled as first order dynamic systems. The inertia of the vehicles is also introduced in the navigation functions to ensure deadlock-free coordination. The proposed approach does not require adaptation of the road infrastructure and relies upon onboard vehicles sensor data. Compared with traffic lights and roundabouts, the proposed method significantly reduces the travel time and the number of stops, thus decreasing energy consumption and pollution emission. This provides a strong motivation to pursue efforts towards the deployment of autonomous vehicles on roads. In the third part of the thesis, we investigate a coordination framework for a large number of miniaturized fiber positioner robots. The fiber positioner robots are designed and built as parts of the next generation of telescopes enabling large spectroscopic surveys. The proposed decentralized framework ensures the collision-free coordination of the fiber positioners sharing a crowed workspace at the focal plate of the telescope. The dynamical (max speed) and the mechanical (limited actuation range) constraints of the positioners are taken into account in the proposed coordination approach, which significantly reduces the time to reach a new robot configuration.

Tagging learning resources in repositories or web portals offers a way to meaningfully describe these resources. The more tags there are, however, the more di cult it is to find one's way around the repository, especially when they are user-generated free-text tags. This paper therefore presents a visualisation of tag clusters based on higher-order co-occurrences that allows users of such repositories a plain but simple way of exploring them in an intuitive manner.

Social media platforms are created and exploited for various activities carried out individually or collaboratively and relying on different resources and tools. Social media platforms are inherently contextual; the context being defined as a specific activity carried out for a specific purpose with specific tools and resources, as well as with specific people. These contexts may be spread over different platforms. Thus, users need to collaborate across various platforms, they need to move their environments and data fromone platformto another. Every task a person accomplishes has its own specifics. Hence, there is a strong need for users to be able to personalize (shape) their environments to suit their specific needs: by changing a set of tools, adding and removing resources, by adapting the graphical and functional parts of their platforms, and sharing resources with others. This thesis investigates the challenges of contextualization, portability and personalization within social media platforms through the following research questions. How can we model a user context in a social media platform? How can we enable portability: i.e., to access the same user's environment from different social media platforms and to migrate an environment from one platformto another? How can we enable the easy personalization of user's contexts? In the first part of the thesis, we formally define the space concept, that materializes the user's context and represents an environment constructed by the user. We propose an OpenSocial space extension that introduces the space concept into OpenSocial specification in the form of Space model and APIs. In addition, we propose a way to build contextual widgets capable of adapting to the user's context. In the second part of the thesis, we propose the notion of collaborative portable space configuration relying on the space configuration language. We demonstrate how portability of spaces can be achieved with OpenSocial. This includes the classification of variousmigration methods and scenarios of space portability. In addition, we propose a concept of portable platforminterfaces. In the third part of the thesis, we define plasticity as a measure of a platform ability to be shaped according to users' needs. To address plasticity, we propose the functional skin concept for personalization of graphical and functional interfaces. In addition, we propose cloud aggregation and sharing mechanisms., Les médias sociaux sont créés et utilisés pour des activités variées menées individuellement ou de manière collaborative et exploitent différents outils et ressources. Les médias sociaux sont intrinsèquement contextuels; le contexte étant défini comme une activité spécifique menée dans un but particulier avec des ressources et des outils choisis, ainsi qu'avec des personnes sélectionnées. Ces contextes peuvent être répartis sur différentes plates-formes que les utilisateurs doivent exploiter conjointement ou entre lesquelles ils doivent échanger leurs données pour collaborer selon leurs besoins spécifiques. Chaque tâche qu'une personne accomplit est unique. Il est donc essentiel pour les utilisateurs de pouvoir personnaliser les espaces qu'ils exploitent dans les médias sociaux en fonction de leurs besoins spécifiques; non seulement en y ajoutant ou supprimant des ressources et en les partageant, mais aussi en modifiant la palette des outils disponibles et en adaptant la visualisation ou les fonctionnalités de l'interface. Cette thèse explore les challenges liés à la contextualisation, à la portabilité et à la personnalisation des médias réseaux en considérant les questions de recherche suivantes: Comment peut-on modéliser un contexte utilisateur dans les médias sociaux? Comment peut-on favoriser la portabilité: c'est-à-dire comment permettre l'accès à un même espace partagé depuis différents médias sociaux ou comment permettre le transfert d'un espace d'un média social à un autre? Comment peut-on permettre une personnalisation aisée d'un espace partagé? Dans la première partie de la thèse, le concept d'espacematérialisant un contexte utilisateur et représentant un environnent personnel est formellement défini. Ensuite, une extension du standard OpenSocial est proposée pour permettre la mise en ouvre du concept d'espace au moyen d'une spécification et d'une API (interface de programmation applicative). Finalement, une manière de construire des applicationsWeb contextuelles comme outils utilisateur s'adaptant au contexte est proposée. Dans la deuxième partie de la thèse, la notion de configuration d'espaces collaboratifs modulables basée sur un langage dédié est introduite. La manière de transférer des espaces partagés grâce à OpenSocial est démontrée. Ceci comprend la classification de différentes méthodes de migration et des scénarii de transfert d'espaces. De plus, le concept d'interface portable indépendante des plates-formes est proposé. Dans la troisième partie de la thèse, la plasticité en tant que mesure de la capacité d'une plate-forme à être personnalisée en fonction des besoins utilisateurs est définie. Ensuite, le concept d'enveloppe de personnalisation graphique et fonctionnelle des interfaces est proposé. Enfin, des solutions simples d'aggrégation de ressources distribuées en nuage et des mécanismes ouverts de partage trans-organisationnels sont développés.

In recent years, research on mash-up technologies for learning environments has gained interest. The overall goal is to enrich or replace traditional learning management systems (LMS) with mash-ups of widgets and services that can be easily combined and configured to fit the learner needs. This paper presents the implemented prototype of the ROLE interoperability framework and a business and an educational case study. The framework provides a common technical infrastructure to assemble widgets and services in Personal Learning Environments (PLEs). Evaluation results indicate that the perceived usefulness and usability is high for one case study in which a mature LMS was enriched with ROLE technology. In the second case study, an early mash-up prototype was deployed. The usefulness and usability of this early prototype were rated low, but the case study provides interesting insights for further research and development. ispartof: pages:125-138 ispartof: Towards Ubiquitous Learning - Proceedings of 6th European Conference of Technology Enhanced Learning, EC-TEL 2011 vol:6964 pages:125-138 ispartof: ECTEL: European Conference of Technology Enhanced Learning location:Palermo, Italy date:20 Sep - 23 Sep 2011 status: published

Two strategies have been proposed in CSCL to foster effective collaboration: structuring the learning scenario by means of collaboration scripts and monitoring interactions among participants in order to detect and regulate potential deviations from the initial plan. In order to help teachers in this endeavor, we propose to combine these two approaches by means of a process where design takes into account the especial requirements posed by monitoring, and monitoring is informed by the characteristics of the scripts that must be met to achieve the learning goals. These desired features are obtained from the constraints defined by the collaborative-learning flow patterns on which the scripts are based. The result is an automated and higher-level view about the evolution of the learning process, integrating the data gathered from the different tools. This paper also presents a case study based on an authentic experience in higher education where these ideas were tried out.

In the framework of mobile augmented reality, a video stream is sent to the user with the help of a wireless communication link. To guarantee an efficient transmission, the video stream rate is controlled by adapting the encoding parameters such as to follow a given bandwidth. The rate can be reduced by reducing the frame rate and/or by choosing a higher compression factor for the video stream. These parameter modifications impact both the level of detail and the fluidity perceived by the user, and thus his/her subjective appreciation. The experience perceived by the user also depends on the context. During a rapid head motion, the notion of fluidity is more important than for a fixed head position. We propose an end-to-end adaptation scheme which enables the encoding of parameters such as to provide the best experience for the user regarding the dynamical context. For example, when the user moves quickly his/her head, the frame is compressed more to increase the frame rate and hence achieve a better perception of the motion. The lack of direct measurement for the subjective user experience is addressed with the design of objective metrics and a generic model to predict the user quality of experience in real time. A rate control strategy based on a systems approach is deployed to manage the multiple encoding parameters which control the stream rate. The encoder is modeled in an abstract manner as a single-variable linear system, where the content variation is taken as a perturbation. A stable and efficient controller is designed for the abstract model of the encoder. To implement the designed controller, the parameter combinations for the real encoder corresponding to the single input of the abstract model should be determined. A new one-pass algorithm determines this correspondence in real time based on a mapping method. Then, the proposed contextual adaptation enables to get the encoding parameter combination that maximizes the quality of experience using an appropriate model. Finally, the global adaptation scheme combines the rate control, the mapping method and the contextual adaptation for real-time implementation. Simulation and experiments illustrate the approach and the global adaptation scheme is validated through different scenarios.

The present project has been carried out in collaboration with a multinational enterprise. The goal is to identify the parameters of an insulin glucose model for Type 1 diabetes. To accomplish this, different population approaches have been implemented to study different dynamical systems and a comparison between them has been done. Furthermore, the algorithm proposed in the study [1] and tested in a previous project [2] has also been examined. These methods allow the identification of the system parameters. The aim is to compare the methods with both artificial and real clinical data. The use of hierarchical models is widely used in pharmaceutical studies where only few observations are available. In this study, it has been proved that the use of the method tested in [2] does not contribute any significant improvement in estimating either the parameters or the model noise. The extension to the second stage has been done with Global Two Stage (GTS), which contributed by substantially improving the duration of the computation time with no loss of precision. This method has proved the effectiveness of the use of hierarchical models for the estimation of fixed-effect parameters as well as the random-effect parameters. Besides, it has been necessary to distinguish the intra and inter individual variation for several subjects from the same experiment. Real clinical data has been tested and the results analyzed to improve the identification process in the near future. Outgoing

Social media platforms are increasingly used in recent years to support learning activities, especially for the construction of activity- and learner-centric personal learning environments (PLEs). This thesis investigates the solutions to four essential design requirements for social media based PLEs: support for help seeking, privacy protection, identity management and activity monitoring, as well as context awareness. Three main components of the thesis, reputation, privacy, and identity, are built upon these four design requirements. We investigate the three components through the following research questions. How do we help learners to find suitable experts or peers who they can learn from or collaborate with in a particular learning context? How can we design a proper privacy mechanism to make sure the information shared by learners is only disclosed to the intended audience in a given context? What identity scheme should be used to preserve the privacy of learners while also providing personalized learning experience, especially for teenage learners? To tackle the design requirement of support for help seeking, we address the reputation dimension in the context of personal learning for doctoral studies, where doctoral students need to find influential experts or peers in a particular domain. We propose an approach to detect a domain-specific community in academic social media platforms. Based on that, we investigate the influence of scholars taking both their academic and social impact into account. We propose a measure called R-Index that aggregates the readership of a scholar's publications to assess her academic impact. Furthermore, we add the social dimension into the influence measure by adopting network centrality metrics in a domain-specific community. Our results show that academic influence and social influence measures do not strongly correlate with each other, which implies that, adding the social dimension could enhance the traditional impact metrics that only take academic influence into account. Moreover, we tackle the privacy dimension of designing a PLE in the context of higher education. To protect against unauthorized access to learners' data, we propose a privacy control approach that allows learners to specify the audience, action, and artifact for their sharing behavior. Then we introduce the notion of privacy protocol with which learners can define fine-grained sharing rules. To provide a usable application of the privacy protocol in social media based PLEs, we exploit the space concept that provides an easy way for users to define the privacy protocols within a particular context. The proposed approach is evaluated through two user studies. The results reveal that learners confirm the usefulness and usability of the privacy enhanced sharing scheme based on spaces. In the last part of the thesis, we study the identity dimension in the context of STEM education at secondary and high schools. To support personalization while also preserving learners' privacy, we propose a classroom-like pseudonymity scheme that allows tracking of learners' activities while keeping their real identities undisclosed. In addition, we present a data storage mechanism called Vault that allows apps to store and exchange data within the scope of a Web-based inquiry learning space.

Recent advances in remote sensing and actuation technologies, coupled with the large reach of the internet, allowed for the emergence of applications such as cyber-physical labs. Cyber-physical labs are the digital and remotely-accessible equivalent of the lab equipment students use in school to experiment, through web-based interfaces such as web applications. Students, teachers and lab owners derive value from these systems, they are our stakeholders. Students are the intended users, teachers are the educational content curators and lab owners are the service providers. In this thesis, we take a close look at issues pertaining to cyber-physical labs and propose new approaches to address them. We also analyze the use of such systems in a MOOC, to detect the impact of the exherted experimental behavior of students on their academic performance. First, we tackle the case of the generation of web apps interfacing cyber-physical labs. It is the equivalent of preparing experiments for teachers by arranging the equipment for multiple experiments with the same equipment. We propose an extension to the Smart Device Specification for cyber-phyiscal labs, and a tool which generates these apps based on it. The automatically generated apps implement the necessary functions to use a cyber-physical lab, and are ready to be integrated in online learning platfroms. Next, we investigate issues related to the collection and retrieval of students' generated data through their interaction with cyber-physical labs. We consider the needs of students and lab owners. Through questionnaires sent to both parties, we elicit the requirements for an activity-tracking infrastructure composed of a vocabulary and an architectural model. The proposed vocabulary ensures deriving value from the recorded activity, and the proposed architecture addresses privacy and data access issues pertaining to students and lab owners respectively. We evaluate our proposal with two example cyber-physical labs. Last, we collect the interaction data with a cyber-physical lab used in a MOOC. We devise computational analyses on the students activity statistics, in search for indicators of academic performance. We find that high and low performing students show statistically different activity statistics. Then, we sequence the steps students did in an experiment, and don't find any statistically significant patterns for low and high-performing students. This analysis provides insights on the usage of installed facilities to service a potential massive access to limited resources (lab installations), and shed light on possible indicators for academic performance.

We live in a rapidly changing digital world marked by technological advances, and fraught with online information constantly growing thanks to the Internet revolution and the online social applications in particular. Formal learning acquired in traditional academic and professional environments is not by itself sufficient to keep up with our information-based society. Instead, more and more focus is granted to lifelong, self-directed, and self-paced learning, acquired intentionally or spontaneously, in environments that are not purposely dedicated for learning. The concept of online Personal Learning Environments (PLEs) refers to the development of platforms that are able to sustain lifelong learning. PLEs require new design paradigms giving learners the opportunity to conduct autonomous activities depending on their interests, and allowing them to appropriate, repurpose and contribute to online content rather than merely consume pre-packaged learning resources. This thesis presents the 3A interaction model, a flexible design model targeting online personal and collaborative environments in general, and PLEs in particular. The model adopts bottom-up social media paradigms in combining social networking with flexible content and activity management features. The proposed model targets both formal and informal interactions where learning in not necessarily an explicit aim but may be a byproduct. It identifies 3 main constructs, namely actors, activities, and assets that can represent interaction and learning contexts in a flexible way. The applicability of the 3A interaction model to design actual PLEs and to deploy them in different learning modalities is demonstrated through usability studies and use-case scenarios. This thesis also addresses the challenge of dealing with information overload and helping end-users find relatively interesting information in open environments such as PLEs where content is not predefined, but is rather constantly added at run time, and differ in subject matter, quality, as well as intended audience. For that purpose, the 3A personalized, contextual, and relation-based recommender system is proposed, and evaluated on two real datasets.

This research work falls on a multidisciplinary domain, where there is a cross feeding between Computer, Information Sciences and Cognitive Sciences. This PhD dissertation has focused on the interaction and collaboration issues in Web-based experimentation. Such context is very complex because (i) while using a Web-based experimentation environment, students can carry out their hands-on activities in a flexible way, (ii) the online learning community is heterogeneous and members have different roles, and (iii) the Web-based experimentation environment itself may integrate a large variety of software components. A framework including a set of models that capture the main features relevant to the complex context of Web-based experimentation in engineering education, the proposed solution and implementation is presented in this dissertation. These models also integrate an extension of the concept of mediation artifacts, which can be experimental results, uploaded documents, or any kind of shared instruments that can be used for supporting and facilitating the online collaborative hands-on activities. Three major aspects in interaction and collaboration that are supported by the mediation artifacts have been addressed, which are the continuity of interaction, the awareness, and the evaluation. First, the notion of continuity of interaction has emerged as an objective that could help users in obtaining a higher quality of interaction and collaboration in a complex environment. The continuity of interaction emphasizes the uninterrupted sequence of activities. I presented the sources of the discontinuity or interaction in such online learning community. I also presented an analysis grid including several dimensions of continuity of interaction. In fact, the artifacts can be exploited to support the interaction between heterogeneous software components constituting the environment; and this process help facilitate the collaboration among users. The provided mechanism is called artifact-based continuity of interaction. Second, I discussed various crucial features of awareness in Web-based experimentation. The artifacts can actually be collected in a public workspace and shared among users. Then the artifacts stored in the workspace can be retrieved, analyzed, and visualized to provide so-called artifact-based awareness that supplies information not only about others' interaction and learning activities but also about the social structure of the whole online learning community. Third, I discussed various evaluation issues related to Web-based experimentation environments. I proposed a model for the evaluation of Web-based experimentation environments, namely the Instrumentation Feedback Model for Evaluation. I also proposed a set of general and artifact-based metrics for measuring various aspects relevant to an online learning community in a Web-based experimentation environment. I also present an extended Web-based laboratory journal for students in engineering curricula, namely eJournal, which is an implementation of the concept of mediation artifacts. The eJournal has been deployed and utilized in many engineering courses at the Ecole Polytechnique Fédérale de Lausanne (EPFL) and other European institutions. Lastly, the dissertation presents an evaluation process carried out from the 2002 winter to the 2005 summer semesters in the automatic control laboratory courses offered by the School of Engineering at the EPFL. The objectives of the evaluation are to study and validate the proposed approach in sustaining interaction and collaboration in the context of Web-based experimentation as well as to improve the utility and usability of the provided environment.

This dissertation studied real-time interaction over the Internet (RTI2) in the context of remote experimentation where there is typically a person at the client side interacting with physical equipments at the server side. Remote experimentations of mechatronic equipment have strong real-time constraints due to the dynamical evolution of the behaviors to be observed and controlled. The objective of this dissertation is to propose solutions that provide the best possible feedback to the user such that the drawbacks inherent to the distance between the client and the remote equipment are minimized while providing sufficient information about the equipment state and its operational conditions. In the ideal solution, the information representing the state of the remote equipment are rendered at the same pace it has been acquired, with a minimal time delay between the acquisition phase and the rendering phase. Solutions are proposed to approach as closely as possible to this ideal solution despite the inherent variability of the Internet and the versatility of the devices used by the clients. To characterize the feedback provided to the user, a definition for the RTI2 Quality of Service is proposed which includes three properties: the level of interaction, the accuracy of dynamic rendering and the wealth of semantic content. As these three properties cannot be directly measured, three related metrics are defined. The metrics definition relies on two abstractions; the first one defines a block as a unit of information that fully describes the state and the conditions of operation of the remote equipment at a given time. The second abstraction defines an end-to-end structure that includes the communication link and both the server and the client devices. The estimated metrics are the block end-to-end round trip time, the block duration ratio and the block size ratio. To guarantee a given quality of service, a cascade end-to-end adaptation structure is proposed. This control scheme is designed to track the block round trip time by adapting the block size. The adaptation scheme and the underlying model are validated with real-world measurements. A mechanism called TCP Most-Recent is proposed as an enhancement to the TCP protocol. It alleviates the problem related to the TCP buffering mechanism, thus improving the data transmission delay. This mechanism is especially useful for applications that send perishable data as in RTI2. Perishable data are data with limited time validity, thus resending this outdated data is useless and inefficient. The tradeoff in using TCP-MR is to accept possible losses in order to gain interactivity. Such a tradeoff can be limited if the adaptation, carried out at the application level, chooses the data to be discarded according to some criteria specified at the user level. This dissertation proposes a broad but consistent foundation for studying RTI2 in the context of remote experimentation. It is build around three main contributions. First, based on the identified objectives of a successful interaction, the quality of service for RTI2 in the context of remote experimentation is defined. The associated metrics and abstractions permitting to work at the right level are also defined. Second, using the provided metrics and abstractions, an adaptation scheme to enforce the QoS provided to the user is proposed and successfully validated. Third, a mechanism called TCP Most-Recent has been proposed to improve the information transmission delay without breaking the Internet best practice.

The economic growth inherent to our nowadays society pushes the industries toward better performances. In the mechatronic context, the increasing competition results in more and more stringent specifications. Thus, the multiple objectives to track become hard to achieve without compromises. A potential interesting solution to this problematic is overactuation, in the sense that, the considered system has more actuated degrees of freedom than the minimal number required to realize a task. Indeed, overactuation enables flexible and efficient responses to a high variety of tasks. Moreover, the coordinated combination of different subsystems enables both to combine their advantages and to cancel their disadvantages. However, the successful coordination of the supplementary degrees of freedom at our disposal, thanks to overactuation, is not trivial. As a matter of fact, the problem of unpredictable response of overactuated systems to a periodic excitation can be particularly critical. Furthermore, the flexibility brought by the overactuation is to be used efficiently in order to justify its corresponding complexity and higher costs. In this sense, the tracking of multiple simultaneous objectives are clearly enabled by the overactuation and thus constitutes a clear motivation for such a solution. As a consequence, the constructive coordination of overactuated systems, which can be very difficult, is very important to achieve stringent objectives. This thesis aims at contributing to the improvement of the coordination of such systems. In this context, three axis of research are considered: differential geometry, potential functions and closed-loop control. Each of these axis is to be taken as a separate insight on the overall coordination of overactuated systems. On the one hand, the formalism of differential geometry enables a solution to the unpredictability problem raised here above. An intelligent parameterization of the solution space to a periodic task enforces the predictability of the subsystem responses. Indeed, the periodicity of the task is transferred to the latter subsystem responses, thanks to an adequate coordination scheme. On the second hand, potential functions enable the coordination of multiple simultaneous objectives to track. A clear hierarchy in the tasks priority is achieved through their successive projections into reduced orthogonal subspaces. Moreover, the previously mentioned predictability problem is also re-examined in this context. Finally, in the frame of an international project in collaboration with the European Southern Observatory (ESO), an opto-mecatronic overactuated system, called Differential Delay Line, enables the consideration of closed-loop coordination. The successful coordination of the subsystems of the Differential Delay Line, combining their intrinsic advantages, is the key control-element ensuring the achievement of the stringent requirements. This thesis demonstrates that a constructive coordination of the supplementary degrees of freedom of overactuated systems enables to achieve, at least partly, the stringent requirements of nowadays mechatronics.

The field of Unmanned Aerial Vehicles (UAVs), also known as drones, is rapidly growing, both in terms of size and of number of applications. Civil applications range from mapping, inspection, search and rescue, taking aerial footage, to art show, entertainment and more. Currently, most applications have a human pilot supervising or controlling the vehicles, but UAVs are expected to gain more autonomy with time. To fly in general airspace, used by both general and commercial aviation, a high level of autonomy is required from UAVs. A core functionality required to fly in general airspace is the UAVs' ability to detect and avoid collisions with other aircraft or objects. This functionality is handled by a so called Sense And Avoid (SAA) system. From among several sensors investigated to be used for a SAA system, a vision-based sensor is seen as a good candidate for a SAA system due to its ability to detect and identify a large variety of objects, as well as being close to the human's main mean to detect aircraft and other objects. To be as general as possible, this work focuses on non-cooperative algorithms that do not take assumptions on the motion of other aircraft. This thesis presents algorithms for a vision-based SAA system. It focuses on the relationship between sensing and avoidance, and how the limitations of one constrain the second. In particular, this thesis studies the consequences of the limited Field Of View (FOV) of a camera sensor on the collision avoidance algorithms. Given the assumptions above, the sensing and tracking of other UAVs is performed using cameras with fish-eye lenses that have a large enough FOV for the collision avoidance algorithms to guarantee to be collision-free. The detection of other UAVs is performed using two methods: a marker-based or a marker-less computer vision algorithms. Using the measurements from the computer vision algorithm, the positions and velocities of neighboring UAVs are tracked using a Gaussian mixture probability hypothesis density filter. This tracking algorithm is able to track multiple UAVs while requiring little computational resources, therefore representing a suitable candidate for on-board deployment. In this work, it is mathematically proven that the motion of an UAV has to be constrained according to the FOV of its sensor. Following that result, several collision avoidance algorithms are adapted to ensure collision-free navigation when used with a sensor with a limited FOV. Sensory limitations such as noise, lag, limited range and FOV, and their effects on the performance of collision avoidance algorithms are studied. Experimental work using high-fidelity simulation and real robots shows that algorithms that only use position information from the sensors are overall more reliable, although less efficient (in terms of distance traveled or trajectory smoothness) than algorithms that also use velocity estimates from the sensing system.

The presence of conversational agents (or chatbots) in educational contexts has been steadily increasing over the past few years. Recent surveys have shown widespread interest in the use of chatbots in education, both for research and practice. Although these surveys highlight tangible benefits and promising future applications of educational chatbots, several challenges limit our ability to integrate these chatbots into educational contexts, including technological, pedagogical, interaction, and design challenges. In this thesis, we motivate our approach to these challenges by formulating one overarching research question: How can we guide the integration of chatbots into domain-specific learning contexts? To address this question and, in turn, tackle some of the challenges identified in the literature, we propose a conceptual framework spanning four different dimensions. Following the design-based research methodology, we address each dimension in a corresponding phase of our design process, undertaking multiple iterations of the design cycle within each phase. Our investigation starts by addressing the technological foundations of our framework through an application development architecture and a learning analytics pipeline aimed at supporting the creation of interactive applications for digital education platforms and providing access to the data generated when learners interact with these applications. Using our architecture, we then zoom in on one domain (software engineering education) and develop the applications needed to scaffold pedagogical scenarios in which these chatbots could interact with learners. Specifically, we propose the code review notebook, a template for building technopedagogical scenarios to support teaching programming best practices. Code review notebooks resemble the interactions developers have on social coding platforms and, given the popularity of chatbots on these platforms, are especially suitable for educational chatbots. In a series of online, observational, and field studies, we then explore different interaction strategies that could be harnessed by educational chatbots in their conversations with learners. In particular, we conducted three field studies to assess the effects that educational chatbots following (i) Wizard of Oz, (ii) rule-based, and (iii) large language model-based conversational strategies could have on different aspects of the learning experience. Findings from these studies are relevant to instructors looking to integrate educational chatbots into their practice and served to inform two final contributions proposed in this thesis. These contributions focus on design processes and comprise a model to guide the participatory design of educational chatbots, as well as a technical blueprint to define how these chatbots could be integrated into digital learning applications. Zooming out from the software engineering education use case, the two final contributions aim to generalize our findings to other educational domains. As application programming interfaces to powerful generative language models become widely accessible, we can only expect that increasingly complex educational chatbots will become ubiquitous in the years to come. Understanding how learners interact with these chatbots and providing the support necessary to guide their development is therefore of paramount importance. Our framework aligns itself closely with this line of research.

This work proposes a new way to design and deploy active multimedia learning environments by non-IT specialists. It aims at defining a model, an architecture and appropriate tools which allow producing learning environments by composing existing learning contents which can be regarded as a combination of multimedia contents and services. The application domain of this work is the Web-based experimentation environments. Nowadays, there are lots of software solutions allowing teachers to produce and deploy Web-based courses simply and efficiently. However, the majority of authoring tools proposed by e-learning solutions are based on models which mainly focus on multimedia content publication and orchestration. Nevertheless, these solutions don't allow managing and orchestrating complex resources like the ones integrated in experimentation environments used at engineers training. To model, produce and deploy learning environments which allow integrating not only multimedia contents but also active resources (tools, services) and orchestrating the whole of them is only possible by specific developments. This work purpose is to propose a new approach making it easy to design, produce and deploy such environments by mixing contents and services. To reach our objective, this work has been organized into four phases. Firstly, we started by studying and analyzing some distance experimentation environments as an example of mixed environments. Based on those observations, the second phase of our work proposed a model and a language allowing the design of active multimedia learning environments by composition and orchestration of existing inert and active learning resources. The third phase was devoted to define an edition and deployment architecture for active multimedia learning environment. Finally, the fourth phase was devoted to propose intuitive and graphical approach in order to produce active multimedia training environments.