Abstract

Robust mapping and localization can be challenging in environments like forests where easily observable landmarks are virtually indistinguishable from each other. In such environments, GPS or robot odometry may also be compromised by the landscape, tree cover, or method of locomotion. Prior work discussed a method for creating geometric hierarchies to describe the relative positions of unlabeled landmarks. Notably, the polygons from these hierarchies were used to perform accurate place recognition from observations containing partial overlap and sensor noise. Here, we use geometric hierarchies to perform robust data association for collections of virtually identical 2D landmarks in direct support of SLAM. Our system, GeoSLAM, utilizes polygon matching to aggregate landmark positions across consecutive observations, reducing the noisiness of input data. We also maintain a global geometric hierarchy of the environment to enable fast and accurate robot pose estimation. Simulation results empirically demonstrate that this system facilitates accurate, real-time localization for robots experiencing significant sensor noise when exploring environments that contain varying densities of indistinguishable landmarks.

Library of Congress Subject Headings

Forests and forestry--Remote sensing; Pattern recognition systems; Robots--Motion; Geometry

Publication Date

7-1-2024

Document Type

Thesis

Student Type

Graduate

Department, Program, or Center

Computer Science, Department of

College

Golisano College of Computing and Information Sciences

Advisor

Zachary Butler

Advisor/Committee Member

Reynold Bailey

Advisor/Committee Member

Fawad Ahmad

Campus

RIT – Main Campus

Plan Codes

COMPSCI-MS

Download

Share

COinS