research
A collection of my research work and academic contributions.
🎯 Research Overview
Bachelor Thesis Research
"Effectiveness of Search-Based Testing on a Deep Reinforcement-Learned Controller"
This thesis represents the culmination of my undergraduate studies in Computer Science at the Technical University of Munich. The work focuses on Search-Based Testing for Deep Reinforcement Learning Controllers, exploring innovative approaches to solve complex computational problems through rigorous research methodology and practical implementation.
🏛️ Institution: Technical University of Munich (TUM)
📅 Completion Date: September 2024
📋 Abstract
This thesis investigates the effectiveness of search-based testing methodologies for evaluating the robustness of deep reinforcement learning controllers. Traditional random testing approaches often fail to uncover critical vulnerabilities in AI systems, particularly in safety-critical applications like autonomous vehicles.
Through empirical evaluation of various metaheuristic search algorithms, this research demonstrates significant improvements in testing efficiency and coverage. The developed OpenSBT framework provides a structured approach for reproducible evaluation of DRL robustness testing methods.
🔬 Methodology
The research employed a comprehensive experimental design comparing multiple search-based testing approaches:
- Genetic Algorithm (GA) - Evolutionary approach for test case generation
- Particle Swarm Optimization (PSO) - Swarm intelligence for exploration
- Simulated Annealing (SA) - Probabilistic optimization technique
- Random Testing - Baseline comparison method
Each algorithm was evaluated across multiple DRL controller architectures and environments, measuring both efficiency and effectiveness metrics.
📊 Key Results
- Conducted empirical evaluation of metaheuristic search algorithms for DRL robustness testing
- Achieved over 500% improvement in testing efficiency compared to traditional random testing
- Developed OpenSBT framework for structured evaluation and reproducibility
- Collaborated with University of Southern California research group for system integration
⚙️ Technical Implementation
The research involved developing sophisticated testing frameworks and evaluation metrics:
🚀 Impact and Future Work
This research contributes to the growing field of AI safety and robustness testing. The OpenSBT framework provides a foundation for future research in automated testing of machine learning systems, particularly in safety-critical applications.
Future work includes extending the framework to support more complex DRL architectures and exploring integration with formal verification methods.