Reinforcement Learning

Reinforcement learning is a type of machine learning that involves an agent learning to make decisions by interacting with an environment. The agent receives feedback in the form of rewards or penalties for its actions, and its goal is to learn a policy that maximizes the cumulative reward over time.

The RL process involves:

1. Exploration: The agent takes actions in the environment to gather information about the consequences of its decisions.
2. Exploitation: The agent uses the gathered information to select actions that maximize the expected reward.
3. Learning: The agent updates its policy based on the rewards or penalties received, with the goal of improving its decision-making over time.

The key insight of RL is that the agent learns through trial and error, without requiring explicit supervision or labeled data. By interacting with the environment and receiving feedback, the agent can learn to make optimal decisions in complex, uncertain situations.

At our institute, research on reinforcement learning primarily focuses on the following goals:

• Sample efficiency: Learning from a small number of experiences, reducing the need for extensive exploration and data collection.
• Interpretability: Understanding the decision-making process of the agent, enabling insights into its behavior and potential biases.
• Unguided exploration: Allowing the agent to discover new strategies and solutions without relying on human-provided guidance or rewards.
• High-level scene representation: Using representations that capture complex scenes, such as those from computer vision (CV) pipelines, to enable the agent to reason about the environment and make informed decisions.

These goals are important for developing reinforcement learning agents that can operate effectively in real-world environments, where data may be limited, and interpretability is crucial for trust and safety. Achieving these goals can lead to more robust, flexible, and autonomous agents.

• Conference Contributions
  • 

    Maximilian Schier, Frederik Schubert, Bodo Rosenhahn

    Explainable Reinforcement Learning via Dynamic Mixture Policies

    2025 IEEE International Conference on Robotics and Automation (ICRA), IEEE, p. To be published, 2025

    (pdfGitHub) BibTeX
  • 

    Tristan Gottwald, Maximilian Schier, Bodo Rosenhahn

    Safe Resetless Reinforcement Learning: Enhancing Training Autonomy with Risk-Averse Agents

    European Conference on Computer Vision Workshops (ECCVW), October 2024

    (pdfGitHub) BibTeX
  • 

    Yannik Mahlau, Frederik Schubert, Bodo Rosenhahn

    Mastering Zero-Shot Interactions in Cooperative and Competitive Simultaneous Games

    Proceedings of the 41st International Conference on Machine Learning (ICML), July 2024

    (arXiv.org) BibTeX
  • 

    Maximilian Schier, Christoph Reinders, Bodo Rosenhahn

    Learned Fourier Bases for Deep Set Feature Extractors in Automotive Reinforcement Learning

    2023 IEEE 26th International Conference on Intelligent Transportation Systems (ITSC), IEEE, September 2023

    (pdfGitHub) BibTeX
  • 

    Maximilian Schier, Christoph Reinders, Bodo Rosenhahn

    Deep Reinforcement Learning for Autonomous Driving Using High-Level Heterogeneous Graph Representations

    2023 IEEE International Conference on Robotics and Automation (ICRA), IEEE, pp. 7147-7153, 2023

    (pdfGithub, Final Paper) BibTeX
  • 

    Carolin Benjamins, Theresa Eimer, Frederik Schubert, André Biedenkapp, Bodo Rosenhahn, Frank Hutter, Marius Lindauer

    CARL: A Benchmark for Contextual and Adaptive Reinforcement Learning

    NeurIPS 2021 Workshop on Ecological Theory of Reinforcement Learning, December 2021

    (arXiv.org) BibTeX
  • 

    Frederik Schubert, Theresa Eimer, Bodo Rosenhahn, Marius Lindauer

    Towards Automatic Risk Adaption in Distributional Reinforcement Learning

    Reinforcement Learning for Real Life (RL4RealLife) Workshop in the 38th International Conference on Machine Learning (ICML), July 2021

    (arXiv.org) BibTeX
• Journals
  • 

    Carolin Benjamins, Theresa Eimer, Frederik Schubert, Aditya Mohan, Sebastian Döhler, André Biedenkapp, Bodo Rosenhahn, Frank Hutter, Marius Lindauer

    Contextualize Me - The Case for Context in Reinforcement Learning

    Transactions on Machine Learning Research, June 2023

    (OpenReview, PDF, arXiv.org) BibTeX
  • 

    Frederik Schubert, Carolin Benjamins, Sebastian Döhler, Bodo Rosenhahn, Marius Lindauer

    POLTER: Policy Trajectory Ensemble Regularization for Unsupervised Reinforcement Learning

    Transactions on Machine Learning Research, April 2023

    (OpenReview, PDF) BibTeX
• Technical Report
  • 

    Frederik Schubert, Theresa Eimer, Bodo Rosenhahn, Marius Lindauer

    Automatic Risk Adaptation in Distributional Reinforcement Learning

    Arxiv Preprint, June 2021

    (arXiv.org) BibTeX