Abstract. In this thesis we investigate the mathematical modeling of cognition using sensorimotor transition systems. The focus of the thesis is enactivism, where an agent learns to think through actions. As a theoretical basis for our implementation, we discuss a mathematical model of enactivist cognition, sensorimotor interaction and how they can be used as algorithmic aides for studying theoretical problems in robotic systems. In Chapter 3 of this thesis, we introduce a platform which was developed in the University of Oulu as a software project and explain how enactivism and sensorimotor interaction have been taken advantage of, in developing a 2D platform. This platform enables one to concretely implement and explore different interaction strategies that allow an agent to construct internal models of its surroundings. The agent in the platform is a multi-jointed robotic arm, which maneuvers through an obstacle-filled environment. The robotic arm tries to explore its environment with minimal sensory feedback, using algorithms created by the user of the platform. Our main goal on this thesis is to implement new features to this platform. We implement a memory functionality which allows the robotic arm to store all its performed actions. The memory helps the agent infer to a greater extent its surroundings from a limited sequence of action-observation pairs, and helps it in getting a better grasp of the environment. In addition, we implement other methods and functionalities, such as an obstacle sensor, a graph visualization of the internal models, etc. to enhance the perceptual ability of the robotic arm. In Section 5, we develop an algorithm for a simple 2D environment with no obstacles. Here the robotic arm makes a 360-degree move in four steps to perceive its surroundings and generates a state machine graph to visualize its internal model of the environment. The goal of the algorithm is to build an accurate representation of the environment with the help of memory. Through this algorithm we are able to ...


    Zugriff

    Download


    Exportieren, teilen und zitieren



    Titel :

    Robotic sensorimotor interaction strategies


    Beteiligte:

    Erscheinungsdatum :

    2023-06-20


    Medientyp :

    Hochschulschrift


    Format :

    Elektronische Ressource


    Sprache :

    Englisch


    Schlagwörter :

    Klassifikation :

    DDC:    629