The goal of RUBIN-VASE is to design and validate novel variational models for the evolution of neuronal activations in the visual and auditory systems, naturally encoding the neurobiological principle of efficient representation. By focusing on modifications of the celebrated Wilson-Cowan equations for neuronal dynamics, we aim to i) validate this as an approach for the primary visual cortex, through the study of hallucinatory patterns; ii) develop a neuro-inspired framework for sound processing and speech reconstruction, by modelling the primary auditory cortex via the same variational evolution; iii) compare the proposed variational models with data-driven ones, based on sparse sensing and predictive coding. Our objectives will follow by coupling the development of rigorous mathematical theories with their numerical and experimental validation. This will be done through an original interaction between variational/control theoretic techniques and psychophyisical experiences.

The general objective of the JCJC research project PANOPLY is to develop a systematic framework for the practical control of networks of linear hyperbolic systems. The proposed control strategies are said to be practical in the sense that they are constructive, easily implementable, and fulfill a given set of performance specifications. Networks of hyperbolic systems, possibly coupled with ordinary differential equations (ODEs), constitute an essential paradigm to describe a wide variety of large complex systems, including wave propagation, traffic network systems, multiscale and multiphysics systems. Controlling and monitoring networks of hyperbolic systems are difficult control engineering problems due to the distributed nature of the different subsystems composing the network (time and space dependency), the possibly involved graph structure of the network, and the physical/economic infeasibility of placing sensors and actuators everywhere along the spatial domain. The stringent operating, environmental and economical requirements and the high mathematical complexity of these systems explain why traditional control methods exhibit a limited range of applicability and have not been successful at high technology readiness levels. Thus, the theory of control of distributed parameter systems needs substantial advancements to achieve control and estimation objectives for such network structures. A network of hyperbolic systems can be described as a graph: each edge corresponds to an elementary hyperbolic subsystem, and interactions between the subsystems composing the network occur at the graph's vertices. This graph representation will be a cornerstone of the methodology we present in PANOPLY. The first objective of PANOPLY is to understand better the links between the network structure (nb of cycles, incidence matrix) and its controllability/observability (C/O) properties. We aim to characterize the configurations for a given graph structure that guarantee C/O. To identify reflections of graph-theoretic notions on the system properties, we will use the concept of structural controllability as a starting point. The second objective is to develop generic analytical techniques to quantify closed-loop performances w.r.t industry-inspired performance indices (e.g., sensitivity, robustness margins, data sampling, convergence rate, computational effort). This objective is crucial to optimize actuators/sensors placement and to tune the controllers we will design. Finally, we aim to design modular, scalable, and numerically implementable output feedback controllers for an admissible configuration of actuators and sensors. The design will introduce degrees of freedom to guarantee potential trade-offs w.r.t implementation constraints. The proposed methodology will be based on a graph representation of the network and its systematic structural analysis. It also relies on the theory of integral equations. We will use recent results obtained by the P.I. for simple networks showing strong relations between spectral controllability and the existence of a solution for a set of appropriate integral Fredholm equations, from which it is possible to derive explicit controllers. We will consider two realistic case studies to validate our theoretical results and compare them to conventional strategies. The first case study corresponds to the problem of traffic control regulation on interconnected freeway segments. It is an academic example we will use to validate our results in simulation for several realistic graph configurations. The second experimental test case is the active control of vibrations in mechanical structures equipped with piezoelectric actuators. This test case will help us analyze the closed-loop performance and the effect of actuator/sensor placement.

Autonomous vehicles, space robots, energy networks, etc., Autonomous Systems (AS) are pervading our society. As AS step up to tackle ever-more complex tasks in unpredictably dynamic situations, they must ensure efficient and reliable operation across a large range of uncertainties. They include hazardous disturbances due to dynamic environments, e.g., pedestrians accidentally crossing or air purification systems inadvertently turning on aboard space outposts. These disturbances are currently scarcely modelled due to their complexity, making provably efficient and safe-against-uncertainty control of AS a high-stake challenge. The urgency is clear: we need trustworthy algorithms to control AS that not only enhance performance but also uphold safety standards against these undermodelled uncertainties. EUtonomous will bridge this gap by moving away from traditional representations of the uncertainty. Thanks to more structured probabilistic models, I will enable accurate modeling of often dangerously undermodelled complex uncertainties through the decision making stack. I will show such refined modeling comes with high rewards: the structure of these models can be leveraged to compute data-driven surrogate (models) that are for the first time proved to be trustworthy in unpredictably dynamic situations. I will then devise novel algorithms for efficient and safe-against-uncertainty control of these surrogates. Theoretical guarantees and fast computations will endow these algorithms with a unique capability to effectively mitigate underrated rare, yet possibly catastrophic outcomes. Although EUtonomous' algorithms will spread across several applications, e.g., autonomous mobility and sustainable energy delivery, I will assess their trustworthiness via experiments on space robots operating in safety-critical circumstances aboard the International Space Station. A major goal consists of testing the autonomy required aboard forthcoming space outposts for Moon colonization.

Ultrafast 3D ultrasound (US) imaging is being developed in research laboratories. However, its clinical application is hindered by insufficient image quality. This project addresses this limitation by proposing various signal and image processing methods for functional 3D US imaging applied to the cardiac muscle. During ischemia, the tissue structure undergoes various changes, such as oedema and reperfusion, and patient management remains under study. Thus, obtaining a local tissue marker related to the orientation of the cardiac tissue is crucial for improving patient management and assessing treatment success. High-quality real-time 3D imaging is necessary to allow practitioners to visualize the heart and perform local measurements. Deep learning algorithms will generate a high-resolution 3D volume from a limited number of low-resolution volumes. This learning process will involve synthetic and experimental acquisitions. Once localized, the anisotropy of the tissue will be measured using the coherence of ultrasound signals. This approach consists of calculating the 3D spatial covariance matrix. Estimating this matrix with limited sample supports compared to the probe elements’ dimension requires specific signal processing techniques to help compute it efficiently, understand it smartly and propose a dedicated estimator of the local anisotropy. Furthermore, the spatial structure of the coherence matrix, derived from the covariance matrix, will be explored. Integrating the evaluation of this coherence matrix as a parameter of a matrix-variable distribution will be studied to adapt to biological signals and ensure stable anisotropy estimation. The project will also address out-of-plane estimation of anisotropy and the extension of the field of view, which is essential for future clinical applications. Currently, measurement is restricted to planes parallel to the probe, limiting its use. In cardiac imaging, a wider field of view is necessary, and the validity of anisotropy measurement with this type of acquisition needs verification. These methodologies will be validated using different in vitro models and an animal ischemia model before evaluation on human subjects. The proposed imaging will be compared with diffusion magnetic resonance imaging, a state-of-the-art technique for estimating cardiac muscle anisotropy but unsuitable for routine clinical use. In conclusion, this project aims to validate the entire acquisition and post-processing chain to develop a new marker of cardiac muscle anisotropy in ultrasound imaging. This marker will be compared with reference imaging techniques to assess its potential for future human clinical applications.

The main purpose of this project is to develop and exploit an all-optical quantum processor, able to chain a succession of operations on light pulses stored in a multi-register quantum memory. This device will allow the implementation of iterative protocols able to efficiently generate arbitrary quantum states of light. This project also includes a theoretical part, with the design of new protocols based on quantum control in order to do the best in terms of fidelity and complexity for the generated states. A main output of this project will be an analysis of the potentialities of such systems, with “realistic” experimental means based on existing or on short-term possible technologies, for efficient generation of GKP states in the optical domain. Such states could open new horizons in quantum information sciences: they would indeed allow an all-optical implementation of quantum gates, which could be an ideal way for scalable quantum computing, where several parts have to be connected.