Ce manuscrit expose notre travail dans le cadre du projet IWARD et détaille la couche de gestion et de décision du groupement de robots. Ce projet avait comme objectif d’assister le personnel médical dans leur travail, ceci est réalisé en utilisant des robots mobiles, reconfigurables, et rechargeables. Ces robots sont conçus pour effectuer des taches logistiques comme : Le transport de médicaments, le nettoyage, le guidage des patients, la surveillance et la téléconsultation. Dans la première partie de la thèse nous présenterons le problème stratégique qui consiste à déterminer les plannings de rechargement des robots, la configuration des robots opérationnels ainsi que la localisation des stations d’attentes des robots lorsqu’ils sont en état de veille. Différentes hiérarchies à plusieurs niveaux de décisions, sont formulées comme des programmes linéaires en nombres entiers. Des formulations utilisant l’approche de génération de colonnes sont aussi développées pour résoudre ces problèmes. Dans la deuxième partie, le problème tactique est exposé, ceci consiste à affecter les taches arrivantes aux différents robots et d’ordonnancer dynamiquement l’exécution ces missions. Deux approches sont inspectées une version centralisée utilisant les algorithmes évolutionnaires et une autre version distribuée utilisant les algorithmes d’enchères inversées. Afin de mettre à l épreuve ces deux approches, une simulation a événements discrets a été conçue et développée spécifiquement pour le projet, permettant ainsi d’évaluer ces deux approches. / Due to the expansion of the life duration and the shortage of medical personal in hospitals the EU funded IWARD project as part of the IFP6 program. The aims of this project were to assist the medical personnel in logistic and non medical tasks (transport, cleaning, environmental monitoring, guidance and tele-monitoring) through the usage of mobile, reconfigurable, rechargeable robots, thus letting the Medical staff to concentrate on medical aspects of their work.This thesis was part of this project, and our work consisted on developing a decision making framework for the team of robots.In the first part of the thesis, we address the strategic decisions essentially the: (i) the robots’ home station location problem, (ii) Robot‘s reconfiguration problems and (iii) Robots recharging scheduling. We formulate those problems as a linear problems and we propose to solve them using Mixed Integer Programming (MIP). We also present a formulation using a column generation approach to solve those problems.In the later part we address the tactical problems, mainly the mission assignment, the mission scheduling and rescheduling. We present two different approaches; a centralized decision finder implemented using genetic algorithms. And a decentralized approach using auction like and market based algorithms in order to provided collaborative decision making framework.Finally we compare those two approaches using a custom made discrete event simulation (DES).

The crew scheduling problem with attendance rates is highly relevant for regional passenger rail transport in Germany. Its major characteristic is that only a certain percentage of trains have to be covered by crew members or conductors, causing a significant increase in complexity. Despite being commonly found in regional transport networks, discussions regarding this issue remain relatively rare in the literature. We propose a novel hybrid column generation approach for a real-world problem in railway passenger transport. To the best of our knowledge, several realistic requirements that are necessary for successful application of generated schedules in practice have been integrated for the first time in this study. A mixed integer programming model is used to solve the master problem, whereas a genetic algorithm is applied for the pricing problem. Several improvement strategies are applied to accelerate the solution process; these strategies are analyzed in detail and are exemplified. The effectiveness of the proposed algorithm is proven by a comprehensive computational study using real-world instances, which are made publicly available. Further we provide real optimality gaps on average less than 10 % based on lower bounds generated by solving an arc flow formulation. The developed approach is successfully used in practice by DB Regio AG.