A Control Theoretic Approach for the Development of an Autonomic Fog Cloud Platform for Vision based Mobile Robot Applications

Abstract

The main contribution of the thesis is to present a control-theoretic approach for the development of an autonomic fog-cloud integrated computing platform for offloading vision data from a mobile robot. We characterize the mobile robot vision data (MRVD) as the variable frame rate and resolution of the objects (under active consideration in a given frame) acquired during navigation. Processing MRVD is computationally intensive and power-hungry, affecting the mobile robot s onboard battery life and navigation performance. Therefore, leveraging an on-premise distributed computing platform such as fog and the cloud for offloading MRVD is a viable solution. newlineTo process the MRVD within a specific time-bound (called the service time) for real-time object detection, deep learning (DL) algorithms are hosted on the constituent compute nodes of a scalable fog-cloud platform. We adopt a control-theoretic approach that comprises developing a linear parameter varying (LPV) framework for modeling and control design using a decentralized LPV-based controller for the fog and the cloud platforms. The configurable parameters available for control design enable modulating the service time with a desirable transient response under time-varying MRVD. We validate the developed theory for a mobile robot while navigating in the application environment such as a warehouse or a shopping mall. The experimental results presented for object detection under service time bounds show the efficacy of the proposed approach. newlineFor localization and trajectory tracking with obstacle avoidance in the application environment, the mobile robot implements an Extended Kalman Filter (EKF) based simultaneous localization and mapping (SLAM) and a bug-based path planning algorithm, respectively. newlineIn addition to the main contribution, the thesis also presents the modeling and model predictive controller (MPC) based control design for navigation of an Omni-directional mobile robot during a single actuator failure. A fault estimation method is used to