Theses and Dissertations

Issuing Body

Mississippi State University

Advisor

Bethel, Cindy L.

Committee Member

Carruth, Daniel W.

Committee Member

Goodin, Chris

Committee Member

Mason, George

Committee Member

Swan, J. Edward, II

Date of Degree

5-13-2022

Document Type

Dissertation - Open Access

Major

Computer Science

Degree Name

Doctor of Philosophy (Ph.D)

College

James Worth Bagley College of Engineering

Department

Department of Computer Science and Engineering

Abstract

Unmanned ground vehicles (UGV) are being explored for use in military domains. Military UGVs operate in complex off-road environments. Vehicle mobility forecasting plays an important role in understanding how and where a vehicle can operate. Traditional mobility forecasting has been done using an analytical model known as the NATO Reference Mobility Model (NRMM). There has been a push to extend the forecasting capabilities of NRMM by integrating more simulation methods. Simulation enables the repeated testing of UGVs in scenarios that would be difficult or dangerous to study in real world testing. To accurately capture UGV performance in simulation, the operating environment must be accurately modeled. Current widely used methods for generating forested virtual environments rely on random methods. These methods result in forests that can appear to be realistic when visually inspected but lack the appropriate distribution of different sizes of vegetation. The size and distribution of vegetation plays a major role in the ability of a vehicle to operate in a forested environment. Therefore, there is a need for alternative forest generation algorithms that generate more realistic virtual forests. To address this, a novel environment generation model based on forest ecology was implemented. This model accurately captures vegetation growth, disbursement, and competition. Simulated UGV self-driving performance for scenes generated using the ecological model was compared to performance for scenes generated using a widely adopted random model. Resulting speeds across each scene were averaged to predict a speed made good (SMG). Vehicle SMG predictions were made in NRMM using scene descriptions matching each of the random and ecological scenes. Using a continuous vegetation override function in simulation, SMG predictions for both methods were similar to the results of NRMM. However, the predicted speeds for scenes generated with the ecological model were different from the predicted speeds for scenes generated with the random model. When examining the distribution and frequency of different sizes of trees, ecological scenes more closely match the distribution and frequency of trees that are expected for real forested environments suggesting that predictions for speed in ecological scenes better represent potential speeds for real environments.

Share

COinS