Novel and relevant technologies are emerging in the field of Mixed Reality (MR). Examples are, Large Language Model-driven agents, or rendering techniques such as Gaussian Splatting. On top of that, security and cybersecurity topics are gaining in importance. In literature reviews, a new perspective on the usage of technologies, comparisons to established ones, and future use cases can lead to a better understanding how the role of MR shifts.

In this project, we aim to investigate pollen-derived molecules that enhance bee fitness and mediate resistance to infection, working with the buff-tailed bumblebee (Bombus terrestris) as the model species. The study assesses how specific nutrients influence, e.g., larval development and survival.

Start: Immediately This project develops a CT-based framework that virtually places a donor liver within a recipient’s abdomen using patient-specific 3D models. It aims to improve transplant matching by identifying anatomical fit problems that conventional size-based methods may miss.

Redesign and build a demonstration device for robot learning from demonstration, taking it from an existing CAD model to a working prototype using CAD and 3D printing. Depending on progress, the device will be used to collect demonstrations and train robot learning models such as VLAs.

In Zusammenarbeit mit einer oberbayerischen Berufsschule soll Deutschunterricht berufsbezogener, kompetenzorientierter und ansprechender gestalten. Im Mittelpunkt stehen die Berufsfelder Bautechnik, Metalltechnik (Handwerk und Industrie) und Agrarwirtschaft.

External Thesis @ Bosch Research Aufgaben - Während deiner Tätigkeit analysierst du den bestehenden Raytracing-basierten Eye-Tracking-Algorithmus und identifizierst dessen rechenintensive Kernfunktionen. - Des Weiteren benchmarkst du den Algorithmus auf einer Mikrocontroller-CPU und bewertest dabei präzise dessen Laufzeit, Speicherbedarf sowie Genauigkeit. - Darüber hinaus untersuchst du detailliert, welche Algorithmus-Teile sich durch den Einsatz von DSP-, Matrix- oder NPU-Einheiten effizient beschleunigen lassen. - Du entwickelst und testest innovative neuronale Modellansätze, um zentrale Eye-Tracking-Berechnungsschritte zu approximieren oder direkt zu beschleunigen. - Die entwickelten Modelle implementierst, quantisierst sowie exportierst du mithilfe von Python/PyTorch und ONNX speziell für die Zielhardware. - Umfassend vergleichst du die unterschiedlichen Ansätze hinsichtlich ihrer Genauigkeit, Laufzeit, des Speicherbedarfs und ihrer Deploybarkeit. - Nicht zuletzt dokumentierst du deine Forschungsergebnisse präzise und leitest daraus konkrete Empfehlungen für die zukünftige Entwicklung ab. Profil - Ausbildung: Studium im Bereich Elektrotechnik, Mechatronik, Embedded Systems, Informatik, Physik, Data Science oder vergleichbar - Erfahrungen und Know-how: Interesse an Mikrocontrollern, Embedded AI, Machine Learning oder Signalverarbeitung; Erfahrung mit Python; Kenntnisse in PyTorch, C/C++, ONNX oder Modellquantisierung sind von Vorteil - Persönlichkeit und Arbeitsweise: du arbeitest gerne praxisorientiert, analysierst systematisch und möchtest Algorithmen auf realer Hardware evaluieren - Arbeitsalltag: hybrides Arbeitsumfeld mit Bürotagen und mobiles Arbeiten nach Absprache; praktische Tests auf Embedded-Hardware sind Teil der Thesis - Sprachen: gutes Deutsch und Englisch Betreuung in Zusammenarbeit mit einem TUM-Lehrstuhl möglich (Chair of Human-Centered Technologies for Learning). Bei Fragen im Voraus kann über den angegebenen Kontakt gerne vermittelt werden. Bei konkretem Interesse bitte direkt über das Bosch Stellenportal bewerben: https://jobs.bosch.de/job/Abschlussarbeit-NPU-Accelerated-AI-for-Eye-Tracking-Algorithm-Development-Embedded-Deployment-Reutlingen-Kusterdingen?id=3c2e23eb-378f-4290-b637-45234e66934a

Can an AI summarize health information without losing what makes it true? TL;DR: This project investigates how well meaning is preserved when large language models generate summaries of medical text, and develops methods to detect information distortion between source documents and generated output.

Start: Immediately If you are interested, please send an informal application including a short statement of motivation (3–4 sentences), your CV, and your transcript of records to tianyu.tang@tum.de. I look forward to receiving your application. (AI-generated application emails will not be considered.)

Robot Learning from Demonstration has promise in laboratory automation tasks that require adaptability and dexterity. Using a multimodal setup that fuses camera, tactile, and proprioceptive data, a 4-DOF SCARA robot arm is trained to perform laboratory manipulation tasks from expert demonstrations, with robustness to variations and disturbances.

Modern VLA models perform well on short‑horizon robotic tasks but often struggle with long‑term planning and multi‑step control. A promising direction is to enhance VLA architectures by integrating them with Physical AI models capable of understanding physical effects, e.g. for challenging, unstructured environments.

How can structured recovery impact employee resilience and wellbeing? A mixed-methods field study on structured recovery effects on health.

We are offering Master’s theses focused on integrating Large Language Models (LLMs) and Vision-Language Models (VLMs) into PCB design workflows using KiCad. The goal is to develop AI agents that automate repetitive engineering tasks, such as datasheet analysis, component placement, and layout validation. Ideal candidates should have strong programming skills (Python/C++), a basic understanding of electronics, and ideally, a background in machine learning.

This project investigates the impact of internal states (e.g. arousal) on visual processing by applying computational and advanced statistical methods.

I have room to supervise about six more Bachelor's and Master's theses starting in the winter semester 2026/27. I am a postdoctoral researcher in the N-Dimensional User Experience group at TUM. My research covers eye tracking, shared attention in virtual and mixed reality, and accessibility in VR and XR. My current open topics are listed on my website at jwgrootjen.com/supervision. Most of them study how people read each other's attention through avatars in VR. Examples include which body parts carry attention cues, how interpersonal distance changes those cues, and whether the findings hold in AR and in the real world. You can take one of these topics as it is. I also encourage you to shape a topic yourself. Before you reach out, read through the open topics and come up with two to three ideas that go in a similar direction. They do not need to be polished. A working title and two or three sentences per idea are enough. We use your ideas as the starting point of our first meeting and pick or refine one of them together. Own ideas in eye tracking, physiological sensing, accessibility, and HCI are welcome as well. To apply, contact me through the contact form on my website using your TUM student email address. Please send a short motivation, your three topic ideas, your transcript of records and, if available, a CV. Several past theses under my supervision grew into publications. The full list of running and completed theses is on the same page.

Cellular Agriculture (Precision Fermentation) - Bewertung proteinreicher Hydrolysate als Nährmedienkomponente in Schüttelkolben- und Bioreaktor-Kultivierungen

In collaboration with World Vision, which implements ecosystem restoration and regeneration projects worldwide, a remote sensing–based monitoring approach will be developed. We are looking for motivated students to support this goal.