Component based design for networked robotics

This thesis presents resultsofmyresearchinthe fieldof programming control applications for robotic tasks. The main aimof thework hasbeen the definition and implementation of a control component framework for robotic applications tha uses a newly proposed middleware for networked embedded systems originated from the researchof European projectRUNES -Reconfigurable Ubiquitous Networked Embedded Systems. Component-based techniques in software development have manybenefits. Components arewell-defined entities that canbe replaced without affecting the rest of the systems, and can thus be developed separately, easily integrated, and are reusable. Such features are crucial for the development of large-scale complex systems. In this context, the component-based middleware developed by theRUNES project offereda flexible architecture for software development for sensor and mobile adhoc networks. Although the middleware already provides components to access network and operating system resources, there has been a need for control-oriented components necessary for the implementation of control applications. The definition, implementation and testing of the control technologies that have been used in the realization of a complex control applicationfor network reconfiguration in a disaster scenario environment will be described in detail. Forthesakeof concreteness, my workhasfocusedondevelopinga control component-based architecture foraspecific application, namely the navigation of autonomous vehicles using primarily vision forlocalization and for building a map of an unknown envirnment. Component-based design helps to separate logic functionalities in subsystems to build a complex architecture, improving reusabilityof code and reconfiguration of the application. Avisual servo scheme is used to steer the wheeled vehicle among locations using reference images. A topological image map is constructed to support this, based on images grabbed bythe on-board camera, along with a global feature-based metric map, obtained using extended Kalman filter techniques, that gives the robot the 3D world perceptions, e.g. world geometry and obstacles. The combination of topological and metric maps allows the use of simple planners and, combined with a robust image processing technique like SIFT, to two important problems in the navigation of robots in unknown environments: the so-called kidnapped robot and closedpath detection problems. The solution here proposed also enables a team of multiple vehicles to merge their information, and to perform navigation using each other’s knowledge of the feasible paths. Experimental results on a laboratory setup are reported, showing the practicality of the proposed approach.