Jannis Blüml
Artificial Intelligence and Machine Learning Group, Computer Science Department, TU Darmstadt
and hessian.AI
S2|02, D216 Hochschulstraße 10, 64289 Darmstadt, Germany
jannis (dot) blueml (at) tu-darmstadt (dot) de
Meetings by appointment.
S2|02, D216 Hochschulstraße 10, 64289 Darmstadt, Germany
jannis (dot) blueml (at) tu-darmstadt (dot) de
Meetings by appointment.
Mission.
My research is centered around reinforcement learning, using deep neural networks. Currently I try to combine modern
approaches in this field with transformer networks as well as develop new approaches. My overall goal is to understand reasoning within RL better,
especially when used in partial observable enviroments.
Availability of thesis topics. If you are not yet familiar with our thesis notes, I recommend you start here: Thesis Proposals at AIML
As I said, my focus is on RL. Accordingly, theses that I supervise often fall into this area. Here it is helpful, but not mandatory, if you are familiar with PyTorch/Tensorflow and Deep Neural Networks. If you already have a topic in this direction in mind or are generally interested, feel free to contact me.
Update 2026: I do not take on any new students for theses or projects at the moment.
Comeback in October 2026.
Availability of thesis topics. If you are not yet familiar with our thesis notes, I recommend you start here: Thesis Proposals at AIML
As I said, my focus is on RL. Accordingly, theses that I supervise often fall into this area. Here it is helpful, but not mandatory, if you are familiar with PyTorch/Tensorflow and Deep Neural Networks. If you already have a topic in this direction in mind or are generally interested, feel free to contact me.
Update 2026: I do not take on any new students for theses or projects at the moment.
Comeback in October 2026.
Project Linktrees
You want to know more about my papers, here are posters, code and everything you need to know.
Representations Matters in Chess (AlphaVile)
Amplify Exploration in MCTS (AMEX)
Causal World Models for Atari (COMET)
From Objects to Skills (Nexus)
DRL Agents are far from humans (HackAtari)
Object-Centric Rewards (OCALM)
Object-Centric Atari Environments (OCAtari)
Object-Centric Attention via Masking (OCCAM)
Timeline.
| 2021 - now: | PhD student at AIML, CS Department, TU Darmstadt, and hessian.AI, Germany |
| 2018 - 2021: | M.Sc. Computer Science at TU Darmstadt, Germany |
| 2014 - 2018: | B.Sc. Computer Science at TU Darmstadt, Germany |
| 2026 | Improving Input Representations in Object-Centric Deep Reinforcment Learning, M.Sc. Thesis |
| 2026 | LLM-Assisted Reward Shaping In Object-Centric Environments, B.Sc. Thesis |
| 2026 | Structured World Understanding: Integrating Object-Centric Learning with Model Based RL, M.Sc. Thesis |
| 2025 | Sustainable City Management through Transformer-based RL with adapotive and boostrapped models, M.Sc. Thesis |
| 2025 | Playing Reconnaissance Blind Chess with AlphaZero, M.Sc. Thesis |
| 2025 | Symbolic Input Representations for DRL playing Atari, B.Sc. Thesis |
| 2025 | Attention-based Reward Shaping in RL via Transformers, M.Sc. Thesis |
| 2024 | LSTM-based beampath optimizer for the S-DALINAC, B.Sc. Thesis |
| 2024 | ISMCTS - Practical Enhancements using Transformers and Theoretical Guarantees for Best Responses, M.Sc. Thesis |
| 2024 | Game Phase Specific Models in A0, M.Sc. Thesis |
| 2023 | Transformers for Urban Energy Management, B.Sc. Thesis |
| 2023 | Using Graph Neural Networks to Improve Generalization in Self-Play Reinforcement Learning, M.Sc. Thesis |
| 2022 | Replacing PUCT with a Planning Model, M.Sc. Thesis |
| 2022 | Multimodal Learning for Chess, B.Sc. Thesis |
| 2022 | Creating an Agent for the Chess Variant Reconnaissance Blind Chess (RBC), M.Sc. Thesis |
| WiSe 2025/26 | JaxTari Lab, Practical Lab, with Quentin Delfosse, Raban Emund, Paul Seitz and Sebastian Wette |
| WiSe 2025/26 | Einführung in die Künstliche Intelligenz, with Prof. Dr. Kristian Kersting, Cedric Derstroff |
| SoSe 2025 | JaxTari Lab, Practical Lab, with Quentin Delfosse |
| WiSe 2024/25 | Arcade 2.0, B.Sc.-Praktikum, with Quentin Delfosse |
| WiSe 2024/25 | Object-centric Reinforcement Learning Lab, Practical Lab, with Quentin Delfosse |
| WiSe 2024/25 | Einführung in die Künstliche Intelligenz, with Prof. Dr. Kristian Kersting, Johannes Czech |
| WiSe 2023/24 | Einführung in die Künstliche Intelligenz, with Prof. Dr. Kristian Kersting, Johannes Czech, Dr. Martin Mundt, |
| WiSe 2023/24 | HackAtari - Object manipulation using the RAM representation in Atari 2600, B.Sc.-Praktikum, with Quentin Delfosse |
| WiSe 2022/23 | Einführung in die Künstliche Intelligenz, with Prof. Dr. Kristian Kersting, Johannes Czech, Dr. Martin Mundt, |
| WiSe 2022/23 | AtariARI 2.0 - Object extraction using the RAM representation in Atari 2600, B.Sc.-Praktikum, with Quentin Delfosse |
Publications
Outputs
–
Latest Year
–
Top Type
–
Loading publications…

