Faculty of Technology: News

The Faculty of Technology warmly invites all interested members of the university community to attend the public presentations as part of the appointment process for the W3 Professorship in Interactive Robotics. The public presentations will take place at the beginning of each candidate interview and offer an opportunity to learn about the candidates’ teaching approaches and current research. Each public session consists of:

• Teaching presentation (in German): 15 minutes, followed by 10 minutes of discussion
• Research presentation (in English): 30 minutes, followed by 15 minutes of discussion

Following the public presentations, the appointment committee will continue with a non-public interview session. We encourage faculty members, researchers, students, and anyone interested in interactive robotics to attend the public presentations and engage in the discussions. 

The interviews are scheduled as follows:

• Karinne Ramirez-Amaro: 29 June, 09:00–12:00, CITEC 1.204
• Rudolf Lioutikov: 30 June, 09:00–12:00, CITEC 1.204
• Rania Rayyes: 30 June, 13:00–16:00, CITEC 1.204
• Julia Starke-Bohse: 30 June, 16:00–19:00, CITEC 1.204
• Tanja Katharina Kaiser: 1 July, 16:00–19:00, CITEC 1.204
• Markus Vorrath: 2 July, 09:00–12:00, CITEC 1.204
• Nicolás Navarro-Guerrero: 6 July, 13:00–16:00, CITEC 1.204
• Niels Dehio: 17 July, 09:00–12:00, CITEC 1.204

To provide further insight into the presentation topics, the following three abstracts are available:

Karinne Ramirez-Amaro

Transparent Robot Decision-Making through Interpretable and Explainable AI

Transparent decision-making enables humans to understand, interpret, and predict what robots do. Interpretable and explainable methods enhance transparency: interpretable methods clarify how a learned model reaches decisions, while explainable methods articulate why specific decisions were made. In this talk, I will first introduce our interpretable AI methods that generate compact, generalizable semantic models to infer human activities, enabling robots to gain a high-level understanding of human movement. Next, I will present our causal approach, which enables robots to anticipate and prevent both imminent and future failures, helping them understand why failures occur, learn from mistakes, and improve future performance. Finally, I will discuss how we combine these methods into a single framework that integrates symbolic planning with hierarchical reinforcement learning. This integration allows us to learn flexible, reusable robot policies for manipulation tasks, yielding coherent action sequences that are individually executable. Together, interpretable and explainable AI form the foundation for general-purpose robots capable of making complex decisions in dynamic and unpredictable environments, and for fostering the mutual understanding that effective human-robot collaboration requires.

Rudolf Lioutikov

Towards Interactive Behavior Foundation Models

The remarkable advances in generative AI have sparked a new wave of robotics research leveraging Diffusion Models and Vision-Language Models, with the ultimate goal of developing Behavior Foundation Models for robotic systems. However, current state-of-the-art approaches face significant limitations that hinder their use in interactive, real-world settings: these models are exceptionally large, resource-intensive, and slow to react, while requiring vast amounts of diverse, well-grounded data to learn spatial and robotic reasoning. Internationally, the push towards Behavior Foundation Models is dominated by ever larger models trained on ever larger datasets, a strategy that leaves the efficiency, reactivity, and grounding required for safe, close human-robot interaction comparatively unexplored. These gaps make it difficult to deploy foundation models that can ground their decisions in cluttered, dynamic scenes, react fluidly to people in real time, and remain transparent in how they behave. This talk presents our recent contributions reducing model size, data requirements, and training/inference costs of Score-Based Diffusion, Vision-Language-Action Models and Movement Primitive based Action Tokenizers, spanning asynchronous architectures for reactive, high-frequency control, scalable pipelines for reliable spatial grounding, and structured, explainable scene representations. Closing these gaps is not merely beneficial but essential for the next generation of robots that must work safely and responsively alongside people, and it gives Germany and Europe a timely opportunity to lead a crucial research direction still neglected by other international players racing to scale Behavior Foundation Models, presenting a path towards Interactive Behavior Foundation Models.

Tanja Katharina Kaiser

Intelligent Multi-Robot Collaboration

Artificial intelligence is enabling robots to become increasingly adaptive, autonomous, and capable of working effectively in teams. In multi-robot systems, AI-based approaches offer new opportunities to improve collaboration across a wide range of settings, from loosely coupled cooperation in shared environments to tightly integrated coordination among heterogeneous robot teams. In this talk, I will cover AI-based approaches to a spectrum of multi-robot collaboration settings, including centralized and decentralized control, homogeneous and heterogeneous teams, and systems with different levels of information sharing. I will discuss cases where robots communicate or share global state information, as well as more challenging scenarios in which each robot must act based only on local observations. The methods I will present are based on different learning methods or combine foundation models with algorithmic approaches, and are exemplified in scenarios relevant for various real-world applications, to enable flexible, scalable, and robust collaboration.