A Modular Framework for CI/CD-Driven Autonomous Vehicle Simulation and Metric Evaluation

This thesis presents the design and implementation of an automated simulation framework for validating the autonomous racing driving software stack, with a focus on reproducibility, scalability, and integration with continuous development workflows. Motivated by operational limitations encountered during real-world testing campaigns, such as manual configuration, limited metric standardization, and time-consuming evaluation, the proposed system introduces a layered architecture composed of input, execution, and output components. At the core of the system lies a GitLab CI/CD-driven orchestration pipeline that enables both manual and event-triggered simulation runs based on YAML-defined scenarios. Each simulation instance is executed in an isolated containerized environment, ensuring consistent behavior across different deployments. A dedicated Metric Analysis Module is responsible for evaluating telemetry against a configurable set of performance metrics, supporting version control, extensibility, and schema validation. The framework integrates with cloud services such as AWS S3 and Lambda to enable fully automated ingestion, processing, and reporting of simulation and real-world driving sessions. Experimental results demonstrate the system’s ability to pre-validate autonomous behaviors, such as overtaking, in simulation prior to physical deployment, significantly reducing human effort while enhancing safety and repeatability. This work contributes a modular, CI-integrated validation infrastructure tailored to the demands of high-frequency testing in autonomous systems development, and lays the foundation for future extensions involving large-scale orchestration and machine learning-driven analysis.