In this project, we explore the use of immersive technologies for collaborative learning. First in fully virtual reality environments and then in a hybrid one which includes different types of devices (e.g. AR/VR, wall displays, desktops), we will design interaction techniques to improve how people collaborate in practical learning activities. We will also develop measures to evaluate the impact of these techniques on how people learn and how it provides users with good social awareness.

Many real-life decision making problems can be modeled as mathematical optimization problems. Often, the data of these problems is not known with precision, because of measurement errors, variability or the time duration of the processes under study, or simple lack of access to reliable data. For instance, in power generation, energy demand is highly uncertain at the time when strategic planning decisions are to be made. In oil production or mining, the measurements concerning the quantity and purity of resources are not sufficiently accurate. Whereas, in a disaster management context, we may simply lack the data to characterize the adverse effects of natural disasters. Robust optimization has evolved as a key paradigm for handling such data uncertainty within mathematical optimization problems: it requires little historical information, can be used without characterizing probability distributions and often leads to tractable optimization problems that can be treated with existing deterministic optimization paradigms. However, the picture is more complex when some of the decisions (referred to as recourse decisions) can be adjusted after the uncertain data is known, to mitigate the effects of uncertainty, leading to adjustable robust optimization problems. Adjustable problems with discrete variables or multiple decision periods are particularly difficult to solve and, up to now, no scalable exact method has emerged. On the other hand, approximate methods, based on restricting the functional form of possible recourse actions, called "decision rules" can be developed. Some examples of decision rules proposed in the literature include affine and piecewise-affine rules in the case where recourse variables are continuous, and piecewise constant rules in the case where some recourse variables are restricted to be integers. The goal of this project is two-fold: on the one hand, we will study novel decision rules in order to improve the quality and generality of applicability of the solutions proposed, on the other hand we will provide numerical quality measures on these solutions by developing corresponding duality tools. This will require studying the theoretical properties of the resulting optimization problems, and developing novel solution algorithms. Our methodological developments will be accompanied by numerical development and testing on applications that include the kidney exchange problem under compatibility uncertainty and the scheduling of nuclear reactors under outage duration uncertainty.

This project is positioned within the general context of computer graphics, focusing on the creation of animated films. Its main goal is to investigate how computer tools can help capturing and reproducing the typicality of traditional 2D animations. More precisely, it plans to address the following challenges: 1) How to interactively generate in-betweening frames from a sparse sequence of rough drawings to swiftly explore different animation choices and motion designs. 2) How to stylize the geometry and motion of a 3D animation to allow its seamless integration into a 2D animation. In both cases, the proposed solutions should be compatible with regular animation workflows and provide adequate controls to artists that should remain at the center of the computer-human interaction loop. Ultimately, those solutions will be integrated inside the same software framework, allowing to produce 2D animations with a unified appearance starting from roughs drawings or 3D inputs.

The field of Mental-Imagery-based Brain-Computer Interfaces (MI-BCIs) has experienced significant growth in recent years but several challenges still need to be overcome to ensure their widespread usability. Even in controlled laboratory settings, the current state of BCI detection accuracy remains suboptimal, i.e., 30% of users cannot use a BCI, remains insufficient in terms of performance, (typically below 75% for two MI classes), shows significant inter-individual variability and often lack intuitiveness and acceptability. To address these challenges (improving performance and usability), my proposal is to design a new BCI paradigm based on somaesthetic stimulations (i.e., which could include neural, muscular and haptic stimuli). Indeed, in preliminary works, I have previously demonstrated that painless stimulation of the median nerve during the mental imagery task, enabled the BCI to be significantly more effective. This opens up the possibility of designing a high-performance, asynchronous BCI with enhanced usability, since the user would be passively stimulated, while brain activity would be monitored via a BCI to deduce if MI was performed or not. My ambition is to build on these very promising results: to develop, study and formalise a new BCI paradigm, exploiting different stimuli for different applications, improving its usability with a user-centred approach and designing new dedicated EEG analysis algorithms. My hypothesis is that when MI coincides with somaesthetic stimulation, the resulting cortical signature is enhanced, making it more detectable by a BCI. In addition to improving BCI performance, this may facilitate the MI task for the user and also improve motor cortex activation, inducing better synaptic plasticity. The research originality is to exploit this phenomenon to enhance MI detection, thereby extending the practical utility of BCIs beyond the confines of the laboratory.