Mechanical Engineering ETDs

Publication Date

Fall 11-14-2016

Abstract

Many robotic applications utilize a detailed map of the world and the algorithm used to produce such a map must take into consideration real-world constraints such as computational and memory costs. Traditional mesh-based environmental mapping algorithms receive data from the sensor, create a mesh surface from the data, and then append the surface to a growing global mesh. These algorithms do not provide a computationally efficient mechanism for reducing redundancies in the global mesh. MABDI is able to leverage the knowledge contained in the global mesh to find the difference between what we expect our sensor to see and what the sensor is actually seeing. This difference between expected and actual allows MABDI to classify the data from the sensor as either data from a novel part of the environment or data from a part of the environment we have already seen before. Using only the novel data, a surface is created and appended to the global mesh. MABDI's algorithmic design identifies redundant information and removes it before it is added to the global mesh. This reduces the amount of memory needed to represent the mesh and also lessens the computational needs to generate mesh elements from the data.

Keywords

Robotics, SLAM, Environmental Mapping, Mapping, RGB-D

Degree Name

Mechanical Engineering

Level of Degree

Masters

Department Name

Mechanical Engineering

First Committee Member (Chair)

Dr. Ron Lumia

Second Committee Member

Dr. Rafael Fierro

Third Committee Member

Dr. Robert Anderson

Document Type

Thesis

Language

English

Download

Share

COinS