SLAM in Hard Places: Evaluating and Developing SLAM for Environments That Break Standard Assumptions Arjun Kumar

Format:

Book

Thesis/Dissertation

Author/Creator:

Kumar, Arjun, author.

Contributor:

University of Pennsylvania. Mechanical Engineering and Applied Mechanics., degree granting institution.

Language:

English

Subjects (All):

0548.

0771.

0984.

Local Subjects:

0548.

0771.

0984.

Physical Description:

1 electronic resource (110 pages)

Contained In:

Dissertations Abstracts International 87-07B

Place of Publication:

Ann Arbor : ProQuest Dissertations and Theses, 2025

Language Note:

English

Summary:

The problem of Simultaneous Localization and Mapping (SLAM) is integral to the deployment of robots in real-world environments. In indoor and highly structured environments, existing approaches that focus on feature extraction and matching are extremely effective given the abundance of distinctive, static features in the workspace that can be detected and registered. In outdoor and marine environments, the absence of distinct features, coupled with changing lighting conditions, amorphous structures, foliage, and natural environmental fluctuations, poses significant challenges to existing SLAM methods, as these conditions violate many of the core assumptions that enable existing SLAM to perform effectively.This work addresses the challenges of performing SLAM in outdoor, natural, and dynamic environments. Specifically, I consider the two fundamental tenets of existing SLAM methods: the existence of distinct environmental features and their static nature. I propose solutions to the SLAM problem when these tenets are no longer valid. To show the limitations of existing SLAM approaches in outdoor natural environments, I evaluate state-of-the-art (SOTA) SLAM in a geological setting where the objective is to map rock formations in the wilderness. In this work, I show how the performance of SOTA SLAM strategies improves when we strategically introduce sparse human-made landmarks into the environment. For environments devoid of static features, such as the deep ocean, I present a novel SLAM formulation that can jointly estimate a dynamics model for the surrounding flow environment and the robot's trajectories. I then deploy this framework for active SLAM and demonstrate how a robot team can select new regions for mapping and exploration that decreases the uncertainty in the resulting map.The main contributions of this work are an evaluation of state-of-the-art SLAM algorithms in an environment devoid of consistent features, the development of a novel SLAM algorithm for use in aquatic feature-free settings, and the development of a novel ASLAM formulation for use in the same setting. By introducing a model of the background flow into the standard SLAM formulation, robots are able improve navigation accuracy and perform uncertainty aware exploration in a dynamic environment characterized by a lack of easily identifiable features

Notes:

Advisors: Hsieh, M. Ani Committee members: Taylor, Camillo J.; Forgoston, Eric; Kumar, Vijay

Source: Dissertations Abstracts International, Volume: 87-07, Section: B.

Ph.D. University of Pennsylvania 2025

Vendor supplied data

Local Notes:

School code: 0175

ISBN:

9798276005393

Access Restriction:

Restricted for use by site license