Constantin Ruhdorfer

Institute for Visualisation and Interactive Systems, University of Stuttgart

  Pfaffenwaldring 5a, 70569 Stuttgart, Germany University of Stuttgart, SimTech Building, Room 01.035 Google Scholar

Biography

Constantin Ruhdorfer is a PhD candidate in the Collaborative Artificial Intelligence group and a scholar of the International Max Planck Research School for Intelligent Systems (IMPRS-IS) since November 2023. He holds a Bachelor’s degree in Computer Science from the Cooperative State University Baden-Württemberg and a Master's degree in Computer Science from the University of Stuttgart. Supervised by Prof. Andreas Bulling, his research focuses on generalisation and adaptation in cooperative AI, spanning multi-agent reinforcement learning, ad-hoc teamwork, and computational Theory of Mind. His long-term goal is to train agents capable of effective human–AI teaming in complex, open-ended environments.

Teaching

2024

Machine Perception and Learning (Tutor)Master

 

Medieninformatik (Teaching Assistant)Bachelor

 

Fachpraktikum -- Practical Course Interactive Systems: Computational Theory of Mind and Cognition (Teaching Assistant)Master

2023

Machine Perception and Learning (Tutor)Master

---

Supervision

2025

AssistUI: Intention-Aware and Machine Theory of Mind Driven Assistive UI Agents (M.Sc.) Busra Balaban

 

Learning to Assist Humans Without Human Data at Scale (M.Sc.) Selim Elbindary

 

Bridging the Synthetic-Experimental Data Gap in the Application of Machine Learning to High-Enthalpy Plasma Spectroscopy (M.Sc.) Marius Krumweide+

 

Model-Based Opponent Modelling as a Pathway to Theory of Mind Reasoning in Hanabi (B.Sc.) Lea Grams

 

Inside the Black Box: Comparing Self-Play and Zero-Shot Coordination Agents in Hanabi (M.Sc.) Philipp Förster**

 

Integration of Machine Learning in High-Enthalpy Plasma Spectroscopy (M.Sc.) Paul Erik Hofmeyer+

2024

Probing Language Models for Theory of Mind in Cooperative Card Games (M.Sc.) Jonas Allali§

 

Leveraging Biologically-plausible Representations for Robust and Efficient Generalization in Reinforcement Learning (M.Sc.) Nan Jiang*

 

Open-Ended Learning via Auto-Curricula for Solving Zero-Shot Theory of Mind Tasks (M.Sc.) Benjamin Caddell

 

The Yokai Challenge: A New Frontier for Multi-Agent Reinforcement Learning and Machine Theory of Mind (M.Sc.) Rui Yang

2023

Evaluation of Different Image Representations for Reinforcement Learning Agents (M.Sc.) Jayakumar Ramasamy Sundararaj*

*co-supervised with Anna Penzkofer §co-supervised with Matteo Bortoletto +co-supervised with Hendrik Burghaus **co-supervised with Fabian Kögel

---

Publications

(Ruhdorfer et al., 2025; Bortoletto et al., 2025; Bortoletto et al., 2025; Ruhdorfer et al., 2025; Ruhdorfer et al., 2025; Hofmeyer et al., 2025; Bortoletto et al., 2024; Bortoletto et al., 2024; Bortoletto et al., 2024; Wang et al., 2024; Ruhdorfer & Schulz, 2020)

1. 

   The Overcooked Generalisation Challenge: Evaluating Cooperation with Novel Partners in Unknown Environments Using Unsupervised Environment Design

   Constantin Ruhdorfer, Matteo Bortoletto, Anna Penzkofer, Andreas Bulling

   Transactions on Machine Learning Research (TMLR), , pp. 1-25, 2025.

   Abstract Links BibTeX Project

   We introduce the Overcooked Generalisation Challenge (OGC) - a new benchmark for evaluating reinforcement learning (RL) agents on their ability to cooperate with unknown partners in unfamiliar environments. Existing work typically evaluated cooperative RL only in their training environment or with their training partners, thus seriously limiting our ability to understand agents’ generalisation capacity - an essential requirement for future collaboration with humans. The OGC extends Overcooked-AI to support dual curriculum design (DCD). It is fully GPU-accelerated, open-source, and integrated into the minimax DCD benchmark suite. Compared to prior DCD benchmarks, where designers manipulate only minimal elements of the environment, OGC introduces a significantly richer design space: full kitchen layouts with multiple objects that require the designer to account for interaction dynamics between agents. We evaluate state-of-the-art DCD algorithms alongside scalable neural architectures and find that current methods fail to produce agents that generalise effectively to novel layouts and unfamiliar partners. Our results indicate that both agents and curriculum designers struggle with the joint challenge of partner and environment generalisation. These findings establish OGC as a demanding testbed for cooperative generalisation and highlight key directions for future research.

   Paper: ruhdorfer25_tmlr.pdf

   Code: https://git.hcics.simtech.uni-stuttgart.de/public-projects/OGC

   @article{ruhdorfer25_tmlr, title = {The Overcooked Generalisation Challenge: Evaluating Cooperation with Novel Partners in Unknown Environments Using Unsupervised Environment Design}, author = {Ruhdorfer, Constantin and Bortoletto, Matteo and Penzkofer, Anna and Bulling, Andreas}, year = {2025}, pages = {1-25}, journal = {Transactions on Machine Learning Research (TMLR)} }

1. 

   ToM-SSI: Evaluating Theory of Mind in Situated Social Interactions

   Matteo Bortoletto, Constantin Ruhdorfer, Andreas Bulling

   Proc. Empirical Methods in Natural Language Processing (EMNLP), 2025.

   Abstract Links BibTeX Project

   Most existing Theory of Mind (ToM) benchmarks for foundation models rely on variations of the Sally-Anne test, offering only a very limited perspective on ToM and neglecting the complexity of human social interactions. To address this gap, we propose ToM-SSI: a new benchmark specifically designed to test ToM capabilities in environments rich with social interactions and spatial dynamics. While current ToM benchmarks are limited to text-only or dyadic interactions, ToM-SSI is multimodal and includes group interactions of up to four agents that communicate and move in situated environments. This unique design allows us to study, for the first time, mixed cooperative-obstructive settings and reasoning about multiple agents’ mental state in parallel, thus capturing a wider range of social cognition than existing benchmarks. Our evaluations reveal that current models’ performance is still severely limited, especially in these new tasks, highlighting critical gaps for future research.

   doi:

   Paper: bortoletto25_emnlp.pdf

   Code: https://git.hcics.simtech.uni-stuttgart.de/public-projects/tom-ssi

   @inproceedings{bortoletto25_emnlp, title = {ToM-SSI: Evaluating Theory of Mind in Situated Social Interactions}, author = {Bortoletto, Matteo and Ruhdorfer, Constantin and Bulling, Andreas}, booktitle = {Proc. Empirical Methods in Natural Language Processing (EMNLP)}, year = {2025}, doi = {} }
2. 

   Brittle Minds, Fixable Activations: Understanding Belief Representations in Language Models

   Matteo Bortoletto, Constantin Ruhdorfer, Lei Shi, Andreas Bulling

   Findings of Empirical Methods in Natural Language Processing (EMNLP), 2025.

   Abstract Links BibTeX Project

   Despite growing interest in Theory of Mind (ToM) tasks for evaluating language models (LMs), little is known about how LMs internally represent mental states of self and others. Understanding these internal mechanisms is critical - not only to move beyond surface-level performance, but also for model alignment and safety, where subtle misattributions of mental states may go undetected in generated outputs. In this work, we present the first systematic investigation of belief representations in LMs by probing models across different scales, training regimens, and prompts - using control tasks to rule out confounds. Our experiments provide evidence that both model size and fine-tuning substantially improve LMs’ internal representations of others’ beliefs, which are structured - not mere by-products of spurious correlations - yet brittle to prompt variations. Crucially, we show that these representations can be strengthened: targeted edits to model activations can correct wrong ToM inferences.

   doi: 10.48550/arXiv.2406.17513

   Paper: bortoletto25_femnlp.pdf

   Code: https://git.hcics.simtech.uni-stuttgart.de/public-projects/mental-states-in-LMs

   @inproceedings{bortoletto25_femnlp, title = {Brittle {{Minds}}, {{Fixable Activations}}: {{Understanding Belief Representations}} in {{Language Models}}}, shorttitle = {Brittle {{Minds}}, {{Fixable Activations}}}, author = {Bortoletto, Matteo and Ruhdorfer, Constantin and Shi, Lei and Bulling, Andreas}, booktitle = {Findings of Empirical Methods in Natural Language Processing (EMNLP)}, year = {2025}, doi = {10.48550/arXiv.2406.17513} }
3. 

   The Yōkai Learning Environment: Tracking Beliefs Over Space and Time

   Constantin Ruhdorfer, Matteo Bortoletto, Andreas Bulling

   IJCAI Workshop on Generative AI & Theory of Mind In Communicating Agents, pp. 1–24, 2025.

   Abstract Links BibTeX Project Oral Presentation

   Developing collaborative AI hinges on Theory of Mind (ToM) - the ability to reason about the beliefs of others to build and maintain common ground. Existing ToM benchmarks, however, are restricted to passive observer settings or lack an assessment of how agents establish and maintain common ground over time. To address these gaps, we introduce the Yokai Learning Environment (YLE) - a multi-agent reinforcement learning (RL) environment based on the cooperative card game Yokai. In the YLE, agents take turns peeking at hidden cards and moving them to form clusters based on colour. Success requires tracking evolving beliefs, remembering past observations, and maintaining common ground with teammates. Our evaluation yields two key findings: First, current RL agents struggle to solve the YLE, even when given access to perfect memory. Second, while belief modelling improves performance, agents are still unable to effectively generalise to unseen partners or form accurate beliefs over longer games, exposing a reliance on brittle conventions rather than robust belief tracking. We use the YLE to investigate research questions in belief modelling, memory, partner generalisation, and scaling to higher-order ToM.

   doi:

   Paper: ruhdorfer25_ijcaiw.pdf

   @inproceedings{ruhdorfer25_ijcaiw, author = {Ruhdorfer, Constantin and Bortoletto, Matteo and Bulling, Andreas}, title = {The Yōkai Learning Environment: Tracking Beliefs Over Space and Time}, booktitle = {IJCAI Workshop on Generative AI & Theory of Mind In Communicating Agents}, year = {2025}, pages = {1--24}, doi = {} }
4. 

   Integration of Machine Learning in High-Enthalpy Plasma Spectroscopy

   Paul Erik Hofmeyer, Hendrik Burghaus, Constantin Ruhdorfer, Johannes Oswald, Andreas Bulling, Georg Herdrich

   International Conference on Flight vehicles, Aerothermodynamics and Re-entry (FAR), pp. 1–8, 2025.

   Abstract Links BibTeX Project

   Optical emission spectroscopy is widely used to characterize high-enthalpy plasmas because it enables measuring a range of plasma parameters. Although the spectra are relatively easy to acquire, extracting meaningful information requires extensive analysis. In this work, a novel approach is developed to automate the analysis of broadband emission spectra by training two machine learning models on synthetic data. The first model is applied to predict plasma temperatures and species number densities in a CO2 plasma jet. The second model is designed to identify the radiation from potentially occurring species in time-resolved spectra of a titanium material sample demising in an air plasma. Developing a synthetic dataset that allows a trained machine learning model to analyze experimental spectra accurately is identified as a major challenge. Overall, these models offer a significant opportunity to automate the analysis of optical emission spectra.

   Paper: hofmeyer25_far.pdf

   @inproceedings{hofmeyer25_far, author = {Hofmeyer, Paul Erik and Burghaus, Hendrik and Ruhdorfer, Constantin and Oswald, Johannes and Bulling, Andreas and Herdrich, Georg}, year = {2025}, pages = {1--8}, title = {Integration of Machine Learning in High-Enthalpy Plasma Spectroscopy}, booktitle = {International Conference on Flight vehicles, Aerothermodynamics and Re-entry (FAR)} }

1. 

   Unsupervised Partner Design Enables Robust Ad-hoc Teamwork

   Constantin Ruhdorfer, Matteo Bortoletto, Victor Oei, Anna Penzkofer, Andreas Bulling

   arXiv:2508.06336, pp. 1-7, 2025.

   Abstract Links BibTeX Project

   We introduce Unsupervised Partner Design (UPD) – a population-free, multi-agent reinforcement learning framework for robust ad-hoc teamwork that adaptively generates training partners without requiring pretrained partners or manual parameter tuning. UPD constructs diverse partners by stochastically mixing an ego agent’s policy with biased random behaviours and scores them using a variancebased learnability metric that prioritises partners near the ego agent’s current learning frontier. We show that UPD can be integrated with unsupervised environment design, resulting in the first method enabling fully unsupervised curricula over both level and partner distributions in a cooperative setting. Through extensive evaluations on Overcooked-AI and the Overcooked Generalisation Challenge, we demonstrate that this dynamic partner curriculum is highly effective: UPD consistently outperforms both population-based and populationfree baselines as well as ablations. In a user study, we further show that UPD achieves higher returns than all baselines and was perceived as significantly more adaptive, more humanlike, a better collaborator, and less frustrating.

   doi: 10.48550/arXiv.2508.06336

   Paper: ruhdorfer25_arxiv.pdf

   @techreport{ruhdorfer25_arxiv, title = {Unsupervised {{Partner Design Enables Robust Ad-hoc Teamwork}}}, author = {Ruhdorfer, Constantin and Bortoletto, Matteo and Oei, Victor and Penzkofer, Anna and Bulling, Andreas}, year = {2025}, pages = {1-7}, doi = {10.48550/arXiv.2508.06336} }

1. 

   VisRecall++: Analysing and Predicting Visualisation Recallability from Gaze Behaviour

   Yao Wang, Yue Jiang, Zhiming Hu, Constantin Ruhdorfer, Mihai Bâce, Andreas Bulling

   Proc. ACM on Human-Computer Interaction (PACM HCI), 8 (ETRA), pp. 1–18, 2024.

   Abstract Links BibTeX Project

   Question answering has recently been proposed as a promising means to assess the recallability of information visualisations. However, prior works are yet to study the link between visually encoding a visualisation in memory and recall performance. To fill this gap, we propose VisRecall++ – a novel 40-participant recallability dataset that contains gaze data on 200 visualisations and five question types, such as identifying the title, and finding extreme values.We measured recallability by asking participants questions after they observed the visualisation for 10 seconds.Our analyses reveal several insights, such as saccade amplitude, number of fixations, and fixation duration significantly differ between high and low recallability groups.Finally, we propose GazeRecallNet – a novel computational method to predict recallability from gaze behaviour that outperforms several baselines on this task.Taken together, our results shed light on assessing recallability from gaze behaviour and inform future work on recallability-based visualisation optimisation.

   doi: 10.1145/3655613

   Paper: wang24_etra.pdf

   Supplementary Material: wang24_etra_sup.pdf

   @article{wang24_etra, title = {VisRecall++: Analysing and Predicting Visualisation Recallability from Gaze Behaviour}, author = {Wang, Yao and Jiang, Yue and Hu, Zhiming and Ruhdorfer, Constantin and Bâce, Mihai and Bulling, Andreas}, year = {2024}, journal = {Proc. ACM on Human-Computer Interaction (PACM HCI)}, pages = {1--18}, volume = {8}, number = {ETRA}, doi = {10.1145/3655613} }

1. 

   Explicit Modelling of Theory of Mind for Belief Prediction in Nonverbal Social Interactions

   Matteo Bortoletto, Constantin Ruhdorfer, Lei Shi, Andreas Bulling

   Proc. 27th European Conference on Artificial Intelligence (ECAI), pp. 866–873, 2024.

   Abstract Links BibTeX Project

   We propose MToMnet - a Theory of Mind (ToM) neural network for predicting beliefs and their dynamics during human social interactions from multimodal input. ToM is key for effective nonverbal human communication and collaboration, yet, existing methods for belief modelling have not included explicit ToM modelling or have typically been limited to one or two modalities. MToMnet encodes contextual cues (scene videos and object locations) and integrates them with person-specific cues (human gaze and body language) in a separate MindNet for each person. Inspired by prior research on social cognition and computational ToM, we propose three different MToMnet variants: two involving fusion of latent representations and one involving re-ranking of classification scores. We evaluate our approach on two challenging real-world datasets, one focusing on belief prediction, while the other examining belief dynamics prediction. Our results demonstrate that MToMnet surpasses existing methods by a large margin while at the same time requiring a significantly smaller number of parameters. Taken together, our method opens up a highly promising direction for future work on artificial intelligent systems that can robustly predict human beliefs from their non-verbal behaviour and, as such, more effectively collaborate with humans.

   doi: 10.3233/FAIA240573

   Paper: bortoletto24_ecai.pdf

   @inproceedings{bortoletto24_ecai, author = {Bortoletto, Matteo and Ruhdorfer, Constantin and Shi, Lei and Bulling, Andreas}, title = {Explicit Modelling of Theory of Mind for Belief Prediction in Nonverbal Social Interactions}, booktitle = {Proc. 27th European Conference on Artificial Intelligence (ECAI)}, year = {2024}, pages = {866--873}, doi = {10.3233/FAIA240573} }
2. 

   Benchmarking Mental State Representations in Language Models

   Matteo Bortoletto, Constantin Ruhdorfer, Lei Shi, Andreas Bulling

   Proc. ICML 2024 Workshop on Mechanistic Interpretability, pp. 1–21, 2024.

   Abstract Links BibTeX Project

   While numerous works have assessed the generative performance of language models (LMs) on tasks requiring Theory of Mind reasoning, research into the models’ internal representation of mental states remains limited. Recent work has used probing to demonstrate that LMs can represent beliefs of themselves and others. However, these claims are accompanied by limited evaluation, making it difficult to assess how mental state representations are affected by model design and training choices. We report an extensive benchmark with various LM types with different model sizes, fine-tuning approaches, and prompt designs to study the robustness of mental state representations and memorisation issues within the probes. Our results show that the quality of models’ internal representations of the beliefs of others increases with model size and, more crucially, with fine-tuning. We are the first to study how prompt variations impact probing performance on theory of mind tasks. We demonstrate that models’ representations are sensitive to prompt variations, even when such variations should be beneficial. Finally, we complement previous activation editing experiments on Theory of Mind tasks and show that it is possible to improve models’ reasoning performance by steering their activations without the need to train any probe.

   doi:

   Paper: bortoletto24_icmlw.pdf

   Code: https://git.hcics.simtech.uni-stuttgart.de/public-projects/mental-states-in-LMs

   Paper Access: https://openreview.net/forum?id=yEwEVoH9Be

   @inproceedings{bortoletto24_icmlw, author = {Bortoletto, Matteo and Ruhdorfer, Constantin and Shi, Lei and Bulling, Andreas}, title = {Benchmarking Mental State Representations in Language Models}, booktitle = {Proc. ICML 2024 Workshop on Mechanistic Interpretability}, year = {2024}, pages = {1--21}, doi = {}, url = {https://openreview.net/forum?id=yEwEVoH9Be} }
3. 

   Limits of Theory of Mind Modelling in Dialogue-Based Collaborative Plan Acquisition

   Matteo Bortoletto, Constantin Ruhdorfer, Adnen Abdessaied, Lei Shi, Andreas Bulling

   Proc. 62nd Annual Meeting of the Association for Computational Linguistics (ACL), pp. 1–16, 2024.

   Abstract Links BibTeX Project

   Recent work on dialogue-based collaborative plan acquisition (CPA) has suggested that Theory of Mind (ToM) modelling can improve missing knowledge prediction in settings with asymmetric skill-sets and knowledge. Although ToM was claimed to be important for effective collaboration, its real impact on this novel task remains under-explored. By representing plans as graphs and by exploiting task-specific constraints we show that, as performance on CPA nearly doubles when predicting one’s own missing knowledge, the improvements due to ToM modelling diminish. This phenomenon persists even when evaluating existing baseline methods. To better understand the relevance of ToM for CPA, we report a principled performance comparison of models with and without ToM features. Results across different models and ablations consistently suggest that learned ToM features are indeed more likely to reflect latent patterns in the data with no perceivable link to ToM. This finding calls for a deeper understanding of the role of ToM in CPA and beyond, as well as new methods for modelling and evaluating mental states in computational collaborative agents.

   doi:

   Paper: bortoletto24_acl.pdf

   Code: https://git.hcics.simtech.uni-stuttgart.de/public-projects/limits-of-tom

   @inproceedings{bortoletto24_acl, author = {Bortoletto, Matteo and Ruhdorfer, Constantin and Abdessaied, Adnen and Shi, Lei and Bulling, Andreas}, title = {Limits of Theory of Mind Modelling in Dialogue-Based Collaborative Plan Acquisition}, booktitle = {Proc. 62nd Annual Meeting of the Association for Computational Linguistics (ACL)}, year = {2024}, pages = {1--16}, doi = {} }

1. 

   Efficient Implementation of Large-Scale Watchlists

   Constantin Ruhdorfer, Stephan Schulz

   PAAR+ SC^2@ IJCAR, pp. 120–133, 2020.

   Links BibTeX Project

   Paper Access: https://ceur-ws.org/Vol-2752/paper9.pdf

   @inproceedings{ruhdorfer20_ijcar, title = {Efficient {Implementation} of {Large}-{Scale} {Watchlists}}, url = {https://ceur-ws.org/Vol-2752/paper9.pdf}, booktitle = {{PAAR}+ {SC}{\textasciicircum}2@ {IJCAR}}, author = {Ruhdorfer, Constantin and Schulz, Stephan}, year = {2020}, pages = {120--133} }