Date of Award
12-2025
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Automotive Engineering
Committee Chair/Advisor
Dr. Rahul Rai
Committee Member
Dr. Venkat Krovi
Committee Member
Dr. Bing Li
Committee Member
Dr. Ruoyu Yang
Abstract
The advancement of autonomous systems, including Connected Autonomous Vehicles (CAVs) and cyber-physical/robotic systems, relies on robust verification and validation (V&V) mechanisms. These processes are essential to ensuring safety, reliability, and consistency, particularly in maintaining predictable performance under diverse conditions. However, existing V&V techniques face challenges such as the complexity of autonomous system data, limited scalability across varying operational scenarios, and the difficulty of identifying and validating edge cases, which are often hard to replicate in controlled environments. Moreover, the dynamic nature of these systems necessitates continuous learning and adaptation, further complicating effective V&V implementation. This dissertation explores V&V aspects in autonomous systems through structured ontology frameworks. These ontologies serve as graphical databases that enable seamless integration with autonomous subsystems, facilitating efficient data organization, verification, and decision-making processes. Two ontology models, the Connected Autonomous Vehicle Ontology (CAVO) and the Scenario Description Language Ontology (SDLO), are developed, each offering distinct applications tailored to specific use cases. The Connected Autonomous Vehicle Ontology (CAVO) addresses interoperability challenges within CAVs. Developed as an extension of the Common Core Ontology (CCO) suite and grounded in the Basic Formal Ontology (BFO), CAVO supports efficient data mapping and retrieval using SPARQL queries. This research emphasizes CAVO’s adaptability to self-driving applications and proposes enhancements to data management for improved efficiency, reliability, and scalability. Additionally, the incorporation of a perception ontology highlights its practical relevance and usability. The Scenario Description Language Ontology (SDLO) streamlines scenario generation in real-time simulation engines using a dedicated natural language interface, allowing stakeholders to specify project requirements with desired fidelity. It converts natural language inputs from a fine-tuned large language model into ontology-compatible triplets, validated through reasoning and Semantic Web Rule Language (SWRL) rules. The Unity simulation engine serves as the real-time platform for this research. An extended scope of this work involves generating multiple scenarios from a single user-defined scenario using SDLO. We propose an automatic scenario generation framework based on a nested constraint-based Bayesian optimization technique. The objective function incorporates newly defined environment-oriented scenario metrics to enrich scenario diversity and realism. We evaluate multiple surrogate models and acquisition functions to determine the optimal configuration for our use case. Initial datasets are created and extended using a Unity-based ontology-driven scenario generation framework. Through iterative experimentation, we identify that a Matern kernel-based Gaussian Process Regressor combined with a q-UCB acquisition function yields the most effective performance for our scenario generation task. Experimental results demonstrate that the proposed approach effectively generates scenario data points that are unique to the environment, while maintaining the integrity of the defined scenario metrics. Furthermore, the method performs competitively compared to baseline objective Bayesian optimization techniques.
Recommended Citation
Poyyamozhi, Shailendran, "Ontology approach for Verification and Validation in Robotics" (2025). All Dissertations. 4156.
https://open.clemson.edu/all_dissertations/4156