Cooperative AI in open-ended environments

I am interested in training agents capable of effective cooperation in complex, open-ended environments.

I am a Ph.D. candidate and research associate in the Collaborative Artificial Intelligence (CAI) group at the University of Stuttgart, advised by Andreas Bulling, and a scholar of the International Max Planck Research School for Intelligent Systems (IMPRS-IS).

My research spans multi-agent reinforcement learning, ad-hoc teamwork, world models, and computational Theory of Mind. I study how agents can generalise to unfamiliar teammates and environments, coordinate without training together, and reason about others' beliefs.

Google Scholar Group profile About me

Research

Generalisation and open-ended learning

How can we train agents to cooperate beyond the partners and tasks they have seen? I develop methods and benchmarks for partner and environment generalisation, including Unsupervised Partner Design and The Overcooked Generalisation Challenge. My recent work also explores multi-agent world models in DreamTeam and continual multi-agent learning in MEAL.

Coordination with unfamiliar partners

How can independently trained agents to cooperate? I study zero-shot coordination through entropy regularisation and grounded joint attention.

Social reasoning and language models

I investigate whether language models can reason about other agents' beliefs and how they represent those beliefs internally, through work such as ToM-SSI and Brittle Minds, Fixable Activations. I am also exploring open-ended and population-based reinforcement learning for safer LLM post-training.

News

HoreKa compute grant

I was awarded a JARDS/NHR Compute Grant of 5,000 GPU hours on the HoreKa supercomputer, approximately €13,000 in compute value.

Paper accepted at NeurIPS 2026

Our paper High Entropy Regularization Leads to Symmetry-Equivariant Policies in Dec-POMDPs has been accepted at NeurIPS 2026. We study how entropy regularisation can improve coordination between independently trained agents.

Yōkai accepted at RLC 2026

Our paper The Yōkai Learning Environment: Tracking Beliefs Over Space and Time has been accepted at RLC 2026. Yōkai is a benchmark for coordination with unfamiliar partners that requires tracking beliefs over space and time.

Two papers accepted at ICML 2026

Our papers Unsupervised Partner Design Enables Robust Ad-hoc Teamwork (oral) and MEAL: A Benchmark for Continual Multi-Agent Reinforcement Learning have been accepted at ICML 2026.

Publications

See Google Scholar for citation information. An asterisk (*) denotes equal contribution.

Preprints

  1. 2026: Benintendi, G.*, Ruhdorfer, C.*, Kögel, F. & Bulling, A. Grounded Joint-Attention Other-Play for Zero-Shot Coordination. Preprint. [arXiv]

Conference and journal papers

  1. 2026: Forkel, J., Ruhdorfer, C., Beukman, M., Bulling, A. & Foerster, J. N. High Entropy Regularization Leads to Symmetry-Equivariant Policies in Dec-POMDPs. NeurIPS 2026. [website] [arXiv]
  2. 2026: Ruhdorfer, C., Bortoletto, M., Forkel, J., Foerster, J. N. & Bulling, A. The Yōkai Learning Environment: Tracking Beliefs Over Space and Time. Reinforcement Learning Conference (RLC). [website] [arXiv]
  3. 2026: Ruhdorfer, C., Bortoletto, M., Oei, V., Penzkofer, A. & Bulling, A. Unsupervised Partner Design Enables Robust Ad-hoc Teamwork. International Conference on Machine Learning (ICML). Oral [website] [arXiv]
  4. 2026: Tomilin, T., van den Boogaard, L., Garcin, S., Ruhdorfer, C., Grooten, B., Kusters, F., Du, Y., Bulling, A., Pechenizkiy, M. & Fang, M. MEAL: A Benchmark for Continual Multi-Agent Reinforcement Learning. International Conference on Machine Learning (ICML). [paper]
  5. 2025: Bortoletto, M., Ruhdorfer, C. & Bulling, A. (2025). ToM-SSI: Evaluating Theory of Mind in Situated Social Interactions. Proc. Empirical Methods in Natural Language Processing (EMNLP) [website] [arXiv]
  6. 2025: Bortoletto, M., Ruhdorfer, C., Shi, L. & Bulling, A. (2025). Brittle Minds, Fixable Activations: Understanding Belief Representations in Language Models. Findings of Empirical Methods in Natural Language Processing (EMNLP Findings) [website] [arXiv]
  7. 2025: Ruhdorfer, C., Bortoletto, M., Penzkofer, A. & Bulling, A. (2025). The Overcooked Generalisation Challenge: Evaluating Cooperation with Novel Partners in Unknown Environments Using Unsupervised Environment Design. Transactions on Machine Learning Research. [paper] [website] [arXiv]
  8. 2024: Bortoletto, M., Ruhdorfer, C., Shi, L. & Bulling, A. (2024). Explicit Modelling of Theory of Mind for Belief Prediction in Nonverbal Social Interactions. Proc. European Conference on Artificial Intelligence (ECAI), pp. 866–873, 2024. (oral) [paper] [website] [arXiv]
  9. 2024: Bortoletto, M., Ruhdorfer, C.*, Abdessaied, A.*, Shi, L. & Bulling, A. (2024). Limits of Theory of Mind Modelling in Dialogue-Based Collaborative Plan Acquisition. Proc. 62nd Annual Meeting of the Association for Computational Linguistics (ACL), pp. 1–16, 2024. [paper] [website] [arXiv] [acl]
  10. 2024: Wang, Y., Jiang, Y., Hu, Z., Ruhdorfer, C., Bâce, M., & Bulling, A. (2024). VisRecall++: Analysing and Predicting Visualisation Recallability from Gaze Behaviour. Proc. ACM on Human-Computer Interaction (PACM HCI), 8(ETRA), 1–18. [paper] [website] [acm]

Workshop papers

  1. 2020: Ruhdorfer, C., & Schulz, S. (2020). Efficient Implementation of Large-Scale Watchlists. PAAR+SC2 @ International Joint Conference on Automated Reasoning (IJCAR) (pp. 120-133). [paper] [website]

Other collaborations

  1. 2025: Hofmeyer, P. E., Burghaus, H., Ruhdorfer, C., Oswald, J., Bulling, A., Herdrich, G. (2025) Integration of Machine Learning in High-Enthalpy Plasma Spectroscopy. Proc. International Conference on Flight vehicles, Aerothermodynamics and Re-entry (FAR), pp. 1–8 [paper] [website]