Robot learning researcher · PhD candidate, KIT
I give robot policies interpretable building blocks: concepts for what they perceive, primitives for how they move.
PhD researcher at the Intuitive Robots Lab, Karlsruhe Institute of Technology, advised by Rudolf Lioutikov, with papers at CoRL, CVPR, ICRA, RSS and NeurIPS. As one of the lab's first PhD researchers, I have helped build it up since its founding in 2022, including its bimanual robot learning stack, from WebXR teleoperation and data processing to policy deployment. In 2025 I interned at Poke & Wiggle, working on MLOps and vision-language-action models for robots.
Research
My research asks how robots can extract richer information from available data rather than simply requiring more of it. Starting from explainable AI, I arrived at the idea that grounding policies in human-understandable concepts can also make them more robust. Under the theme of Concept-Aware Robot Policies (CARP), I investigate how underutilized concepts such as visual affordances and force-aware action representations can improve manipulation.
Perceiving what a scene affords and how it is structured: affordance detection on 3D point clouds, structured image representations, spatial annotations from demonstrations, and photoreal simulated data for robot vision.
Probabilistic movement primitives that take force into account and replan on events, for manipulation tasks where contact matters.
Prototype models that organize the latent space around learned, inspectable prototypes, so each prediction can be explained by comparison to them — including HyperPG, probabilistic prototypes on the hypersphere.
Selected publications
Curriculum vitae
Intuitive Robots Lab, Karlsruhe Institute of Technology. Advised by Rudolf Lioutikov. One of the lab's first PhD researchers, helping build it up since its founding in 2022. Robot learning for manipulation: imitation learning, diffusion policies, affordance detection and force-aware movement primitives.
Poke & Wiggle, Munich. Helped organize and host the Humanoid Manipulation & AI Hackathon at the Deutsches Museum Munich with TUM, and implemented its robot learning stack end to end: raw data processing, policy training and deployment.
SAP, Big Data Intelligence and Research.
Karlsruhe Institute of Technology. Thesis: Imitation Learning for Versatile Robot Manipulation (Autonomous Learning Robots Lab). Research assistant at the Human-Centered Systems Lab and the Autonomous Learning Robots Lab.
TU Darmstadt. Thesis: Development of a Content-Based Image Retrieval System via Deep Learning (Fraunhofer SIT). Exchange student at Tongji University, Shanghai.
Teaching
I co-teach courses on explainable AI, behavior foundation models and robot learning at KIT, and have supervised nine Bachelor's and Master's theses.
Teaching assistantLecture · English
A lecture on the concepts and methods of explainable AI, seen from the perspective of robot learning and human–robot interaction: interpretable scene, task and behavior representations, data-driven and goal-driven explanation, and approaches that bring XAI-based human feedback into the learning process itself.
Students study XAI methods in the context of robot learning in the seminar, then implement and evaluate them in the accompanying research project.
Students study and develop methods for robot learning with behavior foundation models, such as vision-language-action models (VLAs) and WAMs.
An introduction to PyTorch, MuJoCo benchmarks and the tools of robot learning research: students train and evaluate policies in simulation, working with WandB, Hydra and Git.
