Date of Award
8-2026
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Computer Engineering
Committee Chair/Advisor
Dr. Fatemeh Afghah
Committee Member
Dr. Linke Guo
Committee Member
Dr. Adam Hoover
Abstract
Off-road autonomous vehicles (OFF-RAV) have an abundance of potential use-cases, rang- ing from agricultural to safety-critical scenarios. However, modern OFF-RAV development is heavily substantiated by the progress of autonomous systems in the on-road domain. Increasingly complex off-road autonomy and limited domain-specific data have driven the need for exclusive development, but the lack of comprehensive datasets hinders benchmark and model creation. Additionally, prema- ture advancements in synthetic alternatives to real-world data curation and model training through game-engine-hosted simulation platforms have only furthered this gap. The off-road autonomous driving paradigm would thus be greatly aided by advancements in accessible simulation platforms, the availability of a dataset for diverse edge-case training, and the capability of generating realistic natural environments to serve as an effective digital twin. Within this thesis, we address these deficiencies through three individual efforts. First, we create an OFF-RAV simulation platform using Unity game-engine to model dynamic environments for evaluation of model robustness under adverse and evolving conditions. Crucially, we implement a data streaming protocol between two machines that permits practically indefinite data collection, permitting the study of lengthy context windows that better represent real applications. We then use this simulation engine to produce a diverse, 120,000+ dataset across seven different environments. Finally, we utilize Stable Diffusion 2.1 to create a 2.5D terrain generation solution trained on real- world data. By integrating posterior segmentation and texture mapping solutions, we can synthesize an environment within the Unity that is conditioned on location, ecological, and land cover data with acceptable ground-view perspectives.
Recommended Citation
Ross, Timothy D., "An End-to-End Simulation Stack for Off-Road Digital-Twin Development: Integrating Multimodal Data Generation with Generative Terrain Synthesis" (2026). All Theses. 4815.
https://open.clemson.edu/all_theses/4815