The objective of FireFlies is to conduct research on the development of a fleet of heterogeneous small Unmanned Aerial Vehicles (UAVs) to monitor fires. The aim is to collect data on the fire itself and on the associated plume, with little intervention from ground operators. The considered UAVs are mainly heterogeneous from their sensing payload, some being equipped with sensors to remotely observe the fire (IR cameras) and the plume (visible cameras and Lidars), others being devoted to making measurements within the plume (gas and aerosol sensors). FireFlies is an active perception machine that assesses the situation of fires, aiming at endowing civil security forces to predict their short-term evolutions, allowing them to plan countermeasures and prevention actions. FireFlies will develop AI-based methods to (i) map the fire and the associated plume, integrating various tools and formalisms (Gaussian processes regression, 3D surface reconstruction) to fuse the acquired data with fire models, and (ii) actively control the fleet by defining on the basis of the maps of the fire and the plume where, when, and how each UAV should collect data, resorting to reinforcement learning. A lot of effort will be put into experiments, to gather data on small-scale fires and larger-scale plumes, and to assess the efficacy of the fleet control approach in hybrid simulations in which an actual fleet of UAVs explores a simulated fire.

--- Overview Traditional control theorists are concerned with high-level control algorithms and their high-level properties such as convergence, robustness and performance. Notably, they typically assume all calculations are done with real numbers, and do not pay as much attention to the concrete implementation of their control algorithms, or to issues such as precise programming language semantics, or to errors introduced by machine arithmetic. In the CAFEE project, we aim to bridge the gap between control theory and low-level implementations, by providing typical control theory guarantees on the implementation rather than on a high-level algorithm. Overall, the CAFEE research project aims to achieve comprehensive end-to-end verification of control systems, encompassing high-level hybrid models down to the verification of embedded C code, thereby establishing a comprehensive framework for end-to-end verification of control systems. --- Intellectual Merit The CAFEE project will consist of 5 different work-packages. In the first work-package, we will provide an integrated approach for simple linear discrete-time systems, including the design of and end-to-end process achieving verification at all stages, with discrete-time plant semantics. In the second work-package, we will extend the work to focus on hybrid systems consisting of a continuous-time plant dynamics and a discrete-time controller one. While typical control engineers work either with pure continuous-time or pure discrete-time models for verification purpose, the actual system combines both paradigms. We will define a proper semantics model for these hybrid systems, and develop new verification means to reason on these closed loop systems. In the third work-package, we will consider control algorithms that rely on optimization routines, such as model predictive control. Little verification work has been done in this context, and we plan to leverage some of our recent work and develop verification techniques that can be applied to such optimization routines. Our fourth work-package will be an integration task that will focus on numerical accuracy using machine arithmetic, and will integrate the first three work-packages in this context. Verification will then be performed at model level and all along the development cycle, relying on autocoding and target the final hardware platform, considering numerical accuracy of computation. Finally our fifth work-package will apply our techniques on three different applications: car collision avoidance, aircraft collision avoidance, and spacecraft docking. --- Broader Impacts The CAFEE project will build a formal framework enabling full end-to-end verification of control algorithms, with respect to their formal specification. This framework will be released open-source and available to use by academic, government and industrial partners alike. We already work closely with several government agencies (CNES, NASA) and industrial partners (Collins Aerospace, Toyota), and will continue to do so to help with our design. Our government and industrial relationships will ensure technological transfer to those partners. We will also integrate the use of our framework in classes. Another broader impact of the project will be to bring closer together the formal methods community and the traditional control theory community. By bringing together those communities closer, we hope to raise awareness among the control community of such software issues as programming-languages arithmetic and machine precision (both fixed-point and floating-point). Conversely, the formal methods community will learn about the concerns of the controls community, as well as concrete ways to mitigate potential errors, such as using precise robustness measures in the design of control systems. The papers stemming from the project will be published and presented at a combination of formal methods venues and control theory venues.

Our society has entered a data-driven era, in which not only enormous amounts of data are being generated every day, but there are also growing expectations placed on the analysis of these data. OpenData programs, in which data are available for free, are growing in number. A number of popular web sites have opened access to their data through web services in exchange for pecuniary retribution. Analyzing these massive and complex datasets is essential to making new discoveries and creating benefits for people, but it remains a very difficult task. In many cases, the ability to make timely decisions based on available data is crucial to business success, clinical treatments, cyber and national security, and disaster management, but most data have become simply too large and often have too short a lifespan, i.e. it changes too rapidly for classical visualization or analysis methods to be able to handle it properly. The key is not only to visualize data, but also to allow users to interact with the data. This is particularly the case with movement data, such as traffic data on roads or in an airspace, because of the intrinsinc time-dependant nature of these data. Analyzing and understanding time-dependent data poses non-trivial challenges to information visualization. First, such datasets are by their very nature several orders of magnitude larger than static datasets, which underlines the importance of relying on efficient interactions with multiple objects and fast algorithms. Secondly, while patterns of interest in static data can be naturally depicted by specific representations in still visualizations, we do not yet know how to best visualize dynamic patterns, which are inherent to time-dependent data. These are the two challenges that this project aims at addressing. Interaction and representation, with large data, heavily rely on algorithms: algorithms to compute and display the representation, and algorithms to transform the manipulation of the user into updates of the view and the data. Not only do the performance of these algorithms determine what representations can be used in practice, their nature also has a strong influence on what the visualizations look like. The algorithms that are used classiscally in InfoVis are expressed in the data space (e.g. computation on geographic locations). In this project, we will investigate an alternative approach: algorithms expressed in the graphic space (image-based algorithms). This consists of two steps: first, a data representation is built using straightforward InfoVis techniques; second, the resulting image undergoes purely graphical transformations using image processing techniques. This approach takes advantage of changes in the bottlenecks of computer graphics: since data storage and memory limitation is less and less of an issue, we can plan to reduce computation time by using memory as a new tool to solve computationally challenging problems. Furthermore, graphic cards are nowadays used to perform parallel computations (so called GPGPU techniques). We have recently tested this approach to compute static and dynamic bundling of transport flows, and it proved to be a most efficient way of producing representations fast enough to be interactive. This opens a whole field of study, including the scientific validation of the method, its limitations, and its generalization to different types of datasets, other algorithms, and other time-dependent representation patterns. Our goal in this project is to explore new computing techniques with pixel based algorithm to provide efficient visualizations and user interfaces for the exploration of large datasets of time-dependent data. This project theme lies within the basic research category. It is positioned within the areas of Information Visualization, Visual Analytics, Computer Graphics and Human Computer Interaction.

The Knowdget project is part of a larger plan to promote digital devices as empowering tools by improving fundamental knowledge about interaction phenomena and revisiting the architecture of interactive systems. The Knowdget project focuses on widgets, which are building units in toolkits used to create user interfaces. Current widgets have limitations in terms of the actions they support and their discoverability, which hinders the usability of devices for millions of users. The project seeks to redesign widgets, called Knowdgets, to address these limitations, considering multiple user inputs, graphical representation, human capabilities, and information manipulation. The project also explores the definition of programming languages to support the creation of Knowdgets. Preliminary work on Knowdgets, specifically Signifidgets, has started to explore the concept. The project plans to leverage existing non-trivial interactive systems to gather requirements and guide the design of Knowdgets and the supporting software architecture. The project involves the LOKI research team from the Inria centre at the University of Lille and the LII research team at Ecole Nationale de l’Aviation Civile (ENAC).

Mixed-Integer Nonlinear Programming (MINLP) deals with the most general optimization problems, involving both continuous and discrete variables and nonlinear constraint functions. These are among the most challenging computational optimization problems, arising in countless applications from various areas. While research on mixed-integer linear optimization is quite advanced, MINLP is considered an emerging area that is likely to grow in the coming years. MINLP models being in general very difficult to solve, they require exploiting their properties and developing special solution techniques to reduce the computational effort. The ATOMIC project is in the framework of this hot research topic, with the aim of contributing to the advancement in both modeling stimulating real-life problems and developing efficient methods for their solution. A number of challenging problems arising in Air Traffic Management (ATM) constitute interesting research topics particularly in Operations Research and Optimization and naturally lead to MINLP models. Air traffic is at the core of the social and economic dynamism of our society, and an efficient Air Traffic Management has evidently a deep impact on the social, economic, environmental and industrial context. In this framework, a few problems urgently need addressing to ensure a higher level of automation in ATM and consequently more efficiency and safety. The present project focuses on air traffic conflicts, which occur when aircraft are too close to each other according to their predicted trajectories. Mixed-Integer Nonlinear Programming formulations appear to be the natural candidates for these addressed ATM problems, where the need for modeling logical choices suggests the simultaneous presence of mixed (continuous-integer) variables and nonlinear constraints arise from separation condition modeling. Solution algorithms for these ATM problems are mainly based on evolutionary computation. While these methods are computationally efficient, the global optimal solution and even a feasible solution (with no conflict) is not guaranteed to be achieved in a given time. Recent advances in mixed-integer linear and nonlinear programming open new perspectives that have been lacking in earlier researches on conflict avoidance and can have an impact on its effective solution. The present project is therefore aimed to fully exploiting and developing mixed-integer optimization techniques to propose efficient solutions. The optimization will be performed developing specific strategies to deal with the computational difficulty of the target large-scale problems. Deterministic Branch-and-Bound (BB)-type methods (spatial-Branch-and-Bound and interval-Branch-and-Bound variants) will be primarily considered, exploring strategies that can have an impact on the algorithm's ability to provide an optimal solution, including for example strong reformulations and branching strategies. To deal with the difficulty of the problem, other strategies will be also explored, where the optimality guarantee is forsaken in exchange for the computational efficiency. Specifically, we will investigate hybridization of mathematical programming techniques and (meta)-heuristics, in a “matheuristic” framework, where an essential feature is the exploitation of the features of the conceived mathematical programming models of the addressed problem. Starting from the results obtained for the considered specific application, we will seek to identify more general classes of MINLP problems to which the developed techniques can be applied. Expected results of the project include new mathematical models from mixed-integer programming and effective optimization methods, as well as a software library implementing the proposed algorithms.