Master of Engineering (Research) / This thesis deals with the problem of cooperative explorations of a group of autonomous vehicles in unknown environments in the context of decentralized behaviour. The main contribution of this thesis is the development of a comprehensive decentralized cooperative exploration frame work in which each individual vehicle has the ability to explore an unknown environment by itself and also by cooperative behaviour in a team of several vehicles. To simulate the whole system, each individual vehicle will have the ability to explore an unknown environment by dynamically path-planning (with obstacle and collision avoidance), high-level con- trolling, updating the environment map, proposing potential destinations (frontiers), and solving online task assignment. In this thesis, the framework simulates an unknown environment as an occupancy grid map and uses a frontier-base exploration strategy, in which a cell will be marked as a frontier if it is adjacent at least one open cell, as the core architecture. In dealing with the uncertainties in process transition and observation models of autonomous vehicles, the well-known discrete extended Kalman filter (EKF) algorithm is investigated and implemented. When exploring the environment, a vehicle will update its surrounding information, then propose its potential destinations and evaluate the utility (benefit) to travel to each of those destinations. The benefit to go to each destination is derived from the subtraction of the utility (value) of that cell to the sum of the cost to travel to that cell and the steering cost. The key to cooperative exploration in the team of vehicles lies in each vehicle's ability to communicate the updates of the world to the whole team and to contribute to the global list of potential destinations. And each vehicle has the capability of solving the task assignment problem for the team by calling its own online-task-assignment solving engine. This algorithm results each vehicle in having a destination to visit, which benefits the whole team the most and reduces the total exploration time of the team.

Master of Engineering (Research) / This thesis deals with the problem of cooperative explorations of a group of autonomous vehicles in unknown environments in the context of decentralized behaviour. The main contribution of this thesis is the development of a comprehensive decentralized cooperative exploration frame work in which each individual vehicle has the ability to explore an unknown environment by itself and also by cooperative behaviour in a team of several vehicles. To simulate the whole system, each individual vehicle will have the ability to explore an unknown environment by dynamically path-planning (with obstacle and collision avoidance), high-level con- trolling, updating the environment map, proposing potential destinations (frontiers), and solving online task assignment. In this thesis, the framework simulates an unknown environment as an occupancy grid map and uses a frontier-base exploration strategy, in which a cell will be marked as a frontier if it is adjacent at least one open cell, as the core architecture. In dealing with the uncertainties in process transition and observation models of autonomous vehicles, the well-known discrete extended Kalman filter (EKF) algorithm is investigated and implemented. When exploring the environment, a vehicle will update its surrounding information, then propose its potential destinations and evaluate the utility (benefit) to travel to each of those destinations. The benefit to go to each destination is derived from the subtraction of the utility (value) of that cell to the sum of the cost to travel to that cell and the steering cost. The key to cooperative exploration in the team of vehicles lies in each vehicle's ability to communicate the updates of the world to the whole team and to contribute to the global list of potential destinations. And each vehicle has the capability of solving the task assignment problem for the team by calling its own online-task-assignment solving engine. This algorithm results each vehicle in having a destination to visit, which benefits the whole team the most and reduces the total exploration time of the team.

A fundamental problem in mobile robotics is the exploration of unknown fields that might be inaccessible or hostile to humans. Exploration missions of great importance include geological survey, disaster prediction and recovery, and search and rescue. For missions in relatively large regions, mobile sensor networks (MSN) are ideal candidates. The basic idea of MSN is that mobile robots form a sensor network that collects information, meanwhile, the behaviors of the mobile robots adapt to changes in the environment. To design feasible motion patterns and control of MSN, we draw inspiration from biology, where animal groups demonstrate amazingly complex but adaptive collective behaviors to changing environments. The main contributions of this thesis include platform independent mathematical models for the coupled motion-sensing dynamics of MSN and biologically-inspired provably convergent cooperative control and filtering algorithms for MSN exploring unknown scalar fields in both 2D and 3D spaces. We introduce a novel model of behaviors of mobile agents that leads to fundamental theoretical results for evaluating the feasibility and difficulty of exploring a field using MSN. Under this framework, we propose and implement source seeking algorithms using MSN inspired by behaviors of fish schools. To balance the cost and performance in exploration tasks, a switching strategy, which allows the mobile sensing agents to switch between individual and cooperative exploration, is developed. Compared to fixed strategies, the switching strategy brings in more flexibility in engineering design. To reveal the geometry of 3D spaces, we propose a control and sensing co-design for MSN to detect and track a line of curvature on a desired level surface.

Cette thèse présente tout d'abord un méthode d’exploration mono-robot, ensuite une stratégie coopérative décentralisée d'exploration pour un groupe de robots équipés de range finders. Une carte 2-D de la zone explorée est construite sous la forme de figure en pixels et est étendue par les robots en utilisant un planner local qui décide automatiquement entre l'information gagnée et le coût d'exploration. La carte est construite en utilisant la méthode des moindres carrés pour réduire les erreurs des données des capteurs. En divisant la tâche globale en sous-tâches, un contrôleur intelligent permet de réduire la complexité. Cependant, la fusion de différents comportements avec des objectifs différents peut entraîner des contradictions et modifier ainsi la stabilité du système. Par conséquent, la question de mécanisme de coordination de comportements est essentielle pour réaliser un mouvement sécurisé sans collisions. Une méthode intégrée par la coordination des comportements et de commande par fusion est proposée dans le présent travail. Une nouvelle approche basée sur cinq comportements de base pour la navigation de robots mobiles est discutée. Player/Stage est un projet de logiciel open-source pour la recherche sur la robotique. Ses composants comprennent le serveur de réseau et les simulateurs de robot pour plusieurs types de plates-formes de robots. Nous utilisons principalement simulation sous Player/Stage pour tester nos algorithmes en mono-agents/multi-agents , en cartographie et en navigation. Les résultats obtenus montrent que les solutions proposées sont efficaces et peuvent être utilisées dans des robots réels. / In this Ph. D., we will present firstly a single robot exploration method, then a decentralized cooperative exploration strategy for a team of mobile robots equipped with a range finders. A two dimensional map of the explored area is built in the form of a pixel figure. This is expanded by the robots by using a randomized local planner that authomatically realizes a decision between information gain and navigation cost. In our work, the map is reconstructed using a least-mean square method to reduce the errors of the sensor data. In dividing the overall task into subtasks, the intelligent controller allows reducing the robots task complexity. But the fusion of different behaviors with different objectives may cause contradiction in the procedure and alter the stability of the system. Therefore, the issue of behavior coordination mechanisms is crucial in order to realize the non-collision safety-ensured movements. A method integrated by behavior coordination and command fusion is proposed. A new approach with five basic behaviors for mobile robot navigation is discussed.Player/ Stage is an open-source software project for research in robotics and sensor systems. Its components include the Player network server and the Stage robot platform simulators providing a hardware abstraction layer to several popular robot platforms. Player is one of the most popular robot interfaces in research. We mainly use Player/Stage simulation to test our algorithms in mono-agent/multi-agent exploration, map reconstruction and robot navigation. Obtained results show that the proposed approaches are effective and can be applied in real robots.