Date of Award

2024

Document Type

Open Access Dissertation

Degree Name

Doctor of Philosophy in Electrical Engineering (PhD)

Administrative Home Department

Department of Electrical and Computer Engineering

Advisor 1

Jeremy P. Bos

Committee Member 1

Darrell L. Robinette

Committee Member 2

Anthony J. Pinar

Committee Member 3

Michael C. Roggemann

Abstract

In this dissertation, I present a multifaceted study on the generation of artificial terrains using a multifractal method, and their use in autonomous ground vehicle (AGV) planning and navigation performance characterization in high-fidelity simulations, and automated parameter optimization.

My first contribution is a multifractal artificial terrain generation method leveraging the 3D Weierstrass-Mandelbrot function to control terrain roughness. We generate 60 unique off-road terrains while varying the fractal dimension and test the impact on vehicle traversal difficulty. Results show that increasing the fractal dimension decreases low-roughness areas, increases semi-rough and high-roughness areas, and decreases vehicle success rates while increasing vertical accelerations, pitch and roll rates, and traversal times.

My second contribution is an in-depth evaluation of Michigan Technological University's Off-road Navigation Stack (MUONS). MUONS is a terrain-aware navigation approach, leveraging sample-based planning on point clouds. This evaluation involves analyzing results from 30,000 planning and navigation trials in simulation, validated through field testing. Our simulation campaign includes three kinematically challenging terrain maps and twenty combinations of seven path-planning parameters. In simulation, the MUONS-equipped AGV achieved a 0.98 success rate and encountered no failures during field tests. Statistical and correlation analyses identified the Bi-RRT expansion radius used in the initial planning stage as the most correlated with performance metrics such as planning time and traversed path length. Additionally, we observed that the proportional variation due to changes in tuning parameters is highly correlated with field performance. These findings endorse using Monte Carlo simulation campaigns for performance assessment and parameter tuning.

In my third contribution, I address optimizing path-planning parameters for AGVs that navigate off-road environments using point cloud data. This research integrates a high-fidelity simulation environment based on Unreal Engine with Bayesian optimization to automatically tune path-planning parameters. The approach is validated using MUONS. Simulation results demonstrate a significant reduction in the time to plan and traverse from a start to a goal location in an off-road navigation task. The optimized parameters are field tested to ensure applicability beyond the simulated environment, where a 31% (52~s) reduction in mission time over a typical performing set of parameters is found. The findings highlight the pairing of Bayesian optimization and generated terrains in simulation as a valid approach to enhance off-road AGV navigation efficiency and reliability through automated parameter optimization.

Available for download on Friday, August 01, 2025

Included in

Robotics Commons

Share

COinS