Hauptseminar Computer Graphics and Visualization - SS2026

Reinforcement learning: explainability and applications

Advances in reinforcement learning (RL) have led to powerful agents capable of solving complex sequential decision-making problems. At the same time, there is a growing need to better understand, interpret, explain and reliably deploy these systems, especially in real-world settings. Research in explainability, model-based approaches, and practical applications of RL is rapidly evolving, offering new tools to analyze agent behavior, improve sample efficiency, and bridge the gap between simulation and reality.

Fot this seminar each student (or a team of up to three students) must select one specific topic from the list below (pool will be published soon). Topics are organized into topic groups:

Topic Group 1: Explainable Reinforcement Learning

• Converting NN policies to interpretable formats
• Counterfactual explanations for RL agents
• Trajectory summarization and clustering
• Explainability in multi-agent reinforcement learning
• Visualization techniques for reinforcement learning

Topic Group 2: Real-World Applications of Reinforcement Learning

• Applications of RL in a specific domain (e.g., robotics, video games, simulations, LLM, etc)
• Sim-to-real transfer

Topic Group 3: World Models and Model-Based Reinforcement Learning

• State-of-the-art model-based RL methods
• JEPA-like architectures and their application to RL (e.g., TD-JEPA)
• Model predictive control methods for reinforcement learning