Date of Award

5-7-2018

Degree Type

Thesis

Degree Name

M.Sc.

Department

Computer Science

First Advisor

Goodwin, Scott

Keywords

artificial intelligence, graph, navigation, path, pathfinding

Rights

CC-BY-NC-ND

Abstract

Traditional pathfinding techniques are known for calculating the shortest path from a given start point to a designated target point on a directed graph. These techniques, however, are inapplicable to pathfinding problems where the shortest path may prove to be hazardous for traversal, or where multiple costs of differing unit-types lie along the same path. Moreover, the shortest path may not be optimal if it requires forfeiting a valuable resource. While strategic methods have been proposed in the past to completely avoid paths determined to be dangerous, these methods lack the functionality to provide agents the ability to decide which resources are more valuable for conservation, and which resources possess the greatest risk at being lost. For environments where risk varies dynamically across edges, we propose a solution that can determine a path of least expected weight based on multiple properties of edges. With this Multi-Objective Pathfinding technique, agents can make decisions influenced by highest priority objectives and their preferences to trading off some resources for others. The solution is based on traditional pathfinding techniques, extending their usability to cover strategic and dynamic scenarios where additional properties contained within the search map could render them useless. Nevertheless, our solution is compatible with problems where the goal is to simply find the least weighted path, otherwise known as the objectively resource-conservative path among a set of vertices in a graph.

Share

COinS