Dissertation/Thesis Abstract

The Kidnapped Robot Problem as a Classification Problem: Using Artificial Neural Networks to Characterize Robot Localization
by Brennan, Elizabeth, M.S., Saint Louis University, 2019, 275; 13885072
Abstract (Summary)

Autonomous mobile robots in use cases that demand higher levels of position accuracy and update rates than Global Positioning System (GPS) can provide use localization algorithms such as Adaptive Monte Carlo Localization (AMCL) to do rapid online position estimation as they navigate a mapped environment. A localization fault that plagues these mobile robots is known as "The Kidnapped Robot Problem". If a person picks up a robot and puts the robot down somewhere else, the robot must detect that its position has been changed; otherwise, the robot’s operations operate on a faulty pose estimate and the robot will malfunction or be exposed to unexpected and potentially-dangerous situations. Other "kidnapping" conditions that deteriorate the localization pose estimate by causing changes to the robot's driving model include a flat tire or wheel slip on a slick floor. Real-time detection of such kidnapping situations enables existing recovery operations to be initiated so that the robot’s correct pose estimate can be reestablished. The open-source Robot Operating System (ROS) includes an implementation of AMCL that provides localization services for a variety of robot platforms and sensors. While many Kidnap Detection schemes have been designed for specific robot platforms or sensor configurations, this work proposes a novel ROS-based artificial neural network classifier for identifying Robot Kidnapping events by modeling the trends and behavior of the software internals of the Adaptive Monte Carlo Localization algorithm as implemented in the ROS amcl package. This work also developed a novel process for simulating the Kidnapped Robot Problem in the open-source Gazebo robotic simulation environment; this method can be easily customized to simulate other types of robot fault related to robot motion. The developed Kidnap Detection scheme models the probabilistic AMCL algorithm in a light-weight implementation for easy incorporation of Kidnap Detection into ROS robots to improve robot localization operations. Its usability in the ROS environment gives the detection scheme maxiumum reach and impact: as long as the ROS AMCL package is used for localization, the proposed method enables Kidnap Detection operations on a variety of robot platforms and sensors.

Indexing (document details)
Advisor: Mitchell, Kyle
Commitee: Ebel, William, Gururajan, Srikanth
School: Saint Louis University
Department: Engineering
School Location: United States -- Missouri
Source: MAI 58/06M(E), Masters Abstracts International
Source Type: DISSERTATION
Subjects: Computer Engineering, Electrical engineering, Robotics
Keywords: AMCL, Kidnap detection, Probabilistic robotics, ROS, Robot localization, The kidnapped robot problem
Publication Number: 13885072
ISBN: 9781392230886
Copyright © 2019 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy
ProQuest