The work conducted in this thesis contributes to the robotic navigation field by focusing on different machine learning solutions: supervised learning with (deep) neural networks, unsupervised learning, and reinforcement learning. First, we propose a semi-supervised machine learning approach that can dynamically update the robot controller's parameters using situational analysis through feature extraction and unsupervised clustering. The results show that the robot can adapt to the changes in its surroundings, resulting in a thirty percent improvement in navigation speed and stability. Then, we train multiple deep neural networks for estimating the robot's position in the environment using ground truth information provided by a classical localization and mapping approach. We prepare two image-based localization datasets in 3D simulation and compare the results of a traditional multilayer perceptron, a stacked denoising autoencoder, and a convolutional neural network (CNN). The experiment results show that our proposed inception based CNNs without pooling layers perform very well in all the environments. Finally, we propose a two-stage learning framework for visual navigation in which the experience of the agent during exploration of one goal is shared to learn to navigate to other goals. The multi-goal Q-function learns to traverse the environment by using the provided discretized map. Transfer learning is applied to the multi-goal Q-function from a maze structure to a 2D simulator and is finally deployed in a 3D simulator where the robot uses the estimated locations from the position estimator deep CNNs. The results show a significant improvement when multi-goal reinforcement learning is used.


    Zugriff

    Download


    Exportieren, teilen und zitieren



    Titel :

    Machine Learning in Robotic Navigation:Deep Visual Localization and Adaptive Control


    Beteiligte:
    Shantia, Amir (Autor:in)

    Erscheinungsdatum :

    01.01.2021


    Anmerkungen:

    Shantia , A 2021 , ' Machine Learning in Robotic Navigation : Deep Visual Localization and Adaptive Control ' , Doctor of Philosophy , University of Groningen , [Groningen] . https://doi.org/10.33612/diss.157435654



    Medientyp :

    Hochschulschrift


    Format :

    Elektronische Ressource


    Sprache :

    Englisch


    Klassifikation :

    DDC:    629



    Visual robotic behavior learning

    Narayanan, Krishna Kumar / Posada, Luis-Felipe / Hoffmann, Frank et al. | TIB AV-Portal | 2011

    Freier Zugriff

    Situated learning of visual robotic behaviors

    Narayanan, Krishna Kumar / Posada, Luis-Felipe / Hoffmann, Frank et al. | TIB AV-Portal | 2011

    Freier Zugriff

    ROBOTIC ARM LOCALIZATION

    LIU MARK / KO YUNG-CHANG | Europäisches Patentamt | 2024

    Freier Zugriff

    ROBOTIC ARM LOCALIZATION

    LIU MARK / KO YUNG-CHANG | Europäisches Patentamt | 2024

    Freier Zugriff

    Robotic Control with Partial Visual Information

    Kinoshita, K. / Lindenbaum, M. | British Library Online Contents | 2000