The CONTREDO project (Intervals et contractors for dynamic systems) aims at designing a software tool based on intervals to handle dynamic systems. Intervals take naturally into account rounding errors in the computations and bounded uncertainties of the parameters. Intervals thus make it possible to handle in a reliable way dynamic systems for which initial conditions, parameters, control errors and measurements taken along time are known with a bounded uncertainty. Our software tool will propose a general language for defining dynamic systems and original algorithmic operators. The language will enable to define a so-called trajectory constraint satisfaction problem (TCSP), i.e. a system containing ODEs, but also other constraints on the state variables: delay constraints, inter-temporal constraints, non numerical constraints, and so on. The language should allow the manipulation of entities that are not native in standard interval methods, like functions, state variables, i.e. trajectories comprised in a domain forming a "tube", and also variables representing the derivative functions. The CONTREDO project will also work on obtaining efficient solving algorithms. First, the operators defined in the language will be associated to original primitive \textit{contractors} (operators reducing the variable domains in polynomial time). This will enable to contract the domains of the different entities defined in a dynamic system (TCSP), in particular the tubes of the trajectories. Second, this TCSP solver will use new sophisticated contractors to enhance its performance results. Standard guaranteed integration methods, like VNODE-LP, will be encapsulated inside contractors. We will also propose contractors dedicated to specific constraints like delay or inter-temporal constraints. Third, advanced features will be studied to improve the performances, including clever time discretization, bisection or "shaving" interval methods, or symbolic reformulation of expressions (improving interval computations). The project will be validated on several applications. The project feasibility will be first demonstrated on existing maritime robotics applications. This will check the expressiveness power of the language and performance of algorithms. The main application carried out by our industrial partner MBDA deals with the validation and computation of temporal invariance sets (V-stability) and the calculation of an attraction basin. This problem has numerous civil (e.g., autonomous vehicle in urban area, with its security aspects) and military (missile guidance) aeronautic applications. The project will also study the challenging model predictive control problem, which has applications in surgical robotics (adaptive needle insertion). This software tool will be implemented in the free C++ library IBEX (Interval-Based EXplorer) developed by several partners. IBEX is already used by robotic and control practitioners for handling dynamic systems, but no tool currently offers high level features taking into account the dynamics of applications while managing additional constraints and uncertainties on initial conditions and command errors. This is the ambition behind CONTREDO that gathers the main force task who has laid the algorithmic foundations for such a tool.

The VORTEX project proposes a new approach for exploring unknown indoor environments using a fleet of autonomous drones (UAVs). We propose to define a strategy based on swarm intelligence exploiting only vision-based behaviors. The fleet will deploy as a dynamic graph self reconfiguring according to events and discovered areas. Without requiring any mapping or wireless communication, the drones will coordinate by mutual perception and communicate by visual signs. This approach will be developed with RGB and event cameras to achieve fast and low-energy navigation. Performance, swarm properties, and robustness will be evaluated by building a demonstrator extending a quadrotor prototype developed in the consortium.

Fortschritte in der künstlichen Intelligenz haben zur Entwicklung autonomer Agenten (Autos, Boote, Drohnen) geführt, die in einer dynamischen, offenen Umgebung agieren. Wie durch öffentlichkeits\-wirk\-same Unfälle deutlich wurde, bleibt es eine große Herausforderung, deren Sicherheit zu gewährleisten. In diesem Projekt schlagen wir eine zertifizierbare Sicherheitsebene vor, die Entscheidungen im Voraus überwacht und korrigiert. Der Ansatz basiert auf mathematisch rigorosen Techniken aus den formalen Methoden, einer Disziplin der Informatik, die in der Software-Industrie fest etabliert ist und auch in anderen Bereichen, wie z.B. bei cyber-physischen Systemen, immer mehr an Bedeutung gewinnt. Diese Überwachung verhindert nicht nur Unfälle, sondern führt auch zu schnelleren und sichereren Trainingszyklen, da sie zusätzliche Trainingsdaten aus automatisch generierten Familien von kritischen Trajektorien generieren kann. Verwandte Ansätze verwenden Optimierungsverfahren, die aufgrund von Erfüllbarkeitsproblemen, numerischen Fehlern oder hohen Rechenkosten möglicherweise keine gültige Lösung liefern. Im Gegensatz dazu kann unser Ansatz sowohl mathematisch als auch numerisch korrekt und mit vorhersagbaren, niedrigen Laufzeiten implementiert werden. Wir vergleichen unseren Ansatz mit verwandten State-of-the-Art-Ansätzen aus der modellprädiktiven Steuerung und der datenbasierten Vorhersage, indem wir rigorose statistische Tests an realen Systemen durchführen - darunter zwei verschiedene Arten von autonomen Autos, ein Boot und ein Manipulator. Um den Ansatz für zukünftige Generationen von Systemen zugänglich zu machen, bei denen es zunehmend schwieriger wird, Modelle zu erhalten, werden wir untersuchen, inwieweit datenbasierte Ansätze integriert werden können, ohne die Sicherheit und Leistung zu beeinträchtigen. Das vorgeschlagene dreijährige Programm ermöglicht es ENSTA und TUM, ihre wissenschaftliche Expertise und experimentellen Plattformen zu teilen und eine langfristige Zusammenarbeit auf Basis einer gemeinsamen Strategie für vertrauenswürdige Autonomie aufzubauen.

Observing the oceans in coastal and deep offshore zones nowadays relies on coordinated deployments of multiple types of platforms equipped with multiple types of sensors. The ‘multiplatform’ approach is now recognized as the most relevant and cost-effective way to fully describe spatial and temporal oceanic variability for the needs of marine research, ocean observing systems (OOSs) and for the blue economy. Observing and monitoring biological communities (from plankton to fish) is still very challenging, but it is essential to unveil complex ecological processes and ultimately allow adequate marine environmental protection measures and a sustainable exploitation of the ocean. Underwater gliders equipped with novel optical and acoustic imaging sensors have a significant potential to collect and deliver ecosystem data, in particular in extreme environments like the Arctic ocean. Most of the technological building blocks to meet this challenge are available: extremely low power sensors, gliders and software for control and analyses, such as artificial intelligence (AI) algorithms, have been integrated and operated in coordination with other observing platforms, and open new perspectives for comprehensive observations in coastal and deep seas. BIOGLIDER addresses this scientific and technological challenge with an innovative and unique 'bio glider' integrated solution. Three smart devices, a vision profiler, a scientific echosounder and an acoustic modem will be integrated on commerciallyavailable gliders to provide a ‘smart’ service for zooplankton and fish ecology applications. It will be tested in Nordic seas and the Arctic ocean, meeting the needs of a wide range of customers, from research to the energy and fishery sectors. BIOGLIDER will develop this innovative marine technology expertise in Europe through a strong, organized public-private collaboration, leading to the only commercialized solution for a glider-based ecosystem payload available worldwide.

The TAURUS project (Traversability analysis for AUtonomous Robot and Unmanned System) brings together EXAIL Robotics and ENSTA Paris to answer the MOBILEX challenge on navigation in unstructured environments. EXAIL Robotics brings its expertise in the design of robust and operational autonomous systems in order to design and build a proof of concept meeting all the constraints of the challenge. In addition, it brings its expertise in the remote operation of mobile platforms in unstructured terrain. ENSTA Paris brings its expertise in the field of environmental perception, machine learning and autonomous navigation using LIDAR and visual sensors to carry out the missions proposed in the challenge. The approach adopted by the project follows an incremental progression integrating in the first year existing software bricks in the state of the art or from partners that will be developed to meet the requirements of the following years' challenges. The first iteration will allow the hardware design with the integration of LIDAR, stereovision, infrared, GPS and IMU sensors and LIDAR navigation bricks using a geometric approach, visual navigation with a self-supervised approach to traversability, robust localisation integrating GPS, IMU, LIDAR and visual SLAM. These approaches will be extended by the development of new learning approaches using LIDAR, visual and infrared data to deal with the most complex situations, as well as a more efficient navigation approach using MPC (Model Predictive Control). This project will advance the state of the art in terms of knowledge of the performances achievable by the different approaches integrated and tested, as well as by making available to the community databases representative of the problems addressed using the Barracuda platform. It will also propose new methods for processing and fusing LIDAR, visual and infrared data for autonomous navigation in complex situations, as well as a new approach for fast MPC control based on a 3D semantic map and machine learning methods. Finally, it will enable the further development of robust navigation building blocks that can be rapidly integrated into new products or projects. The TAURUS project will result in the hardware/software design of a proof of concept that will be evaluated during the three challenges of the MOBILEX challenge.