We regularly offer bachelor and masterthesis projects for motivated students who are excited to explore research questions in our areas of interest. Topics vary depending on ongoing projects, but we are always open to new ideas and collaborative exploration. Below is a selection of current or recent thesis topics to give you a sense of what working with us might look like.

Process

Research Areas: Topics (subject to availability) are offered in:
Reinforcement Learning, Program Synthesis, Computer Vision, Mechanistic Interpretability, GUI Prototyping, Explainability, Autonomous Driving.
To apply, send your topic of interest, Transcript of Records, and CV to patrick.knab@tu-clausthal.de.
Writing Expose: If a topic is available, the supervising researcher will help define a concrete topic. You must then submit an expose including:
- Background & Motivation
- Goals of the Thesis
- Work Plan (at least 2 pages, excl. references)
Once accepted, you may begin your thesis project.
Thesis Project: Maintain regular meetings with your supervisor and make consistent progress. Use the provided LaTeX template and write in English. Focus your report on your results; use the appendix for supplementary materials.
Submission: Submit 2 printed copies of your report and 2 CDs/USB sticks with code, data, and installation instructions to the secretary's office (Wallstr. 6, Goslar). Arrange an appointment in advance and clarify additional requirements with your supervisor.

Topics

Summer Semester 2025

This semester, we offer the following topics. If not stated otherwise, the topics are open to bachelor and master students.

Understanding and Mitigating Goal Misgeneralisation in Neural Networks. Goal misgeneralisation occurs when a neural network performs well during training but optimizes for the wrong objective when deployed in new settings. This project will use tools from mechanistic interpretability to reverse-engineer how specific internal components (e.g., neurons, attention heads) encode and act upon implicit goals in neural models. By identifying the circuits and representations responsible for goal-directed behavior and their failure modes, we aim to propose targeted interventions that reduce misalignment. The findings could inform the design of safer, more reliable AI systems that generalize their goals more robustly.
NLP/LLM-Driven Requirements Engineering and GUI Prototyping. In this area the focus lies on integrating Natural Language Processing (NLP) techniques and Large Language Models (LLMs) into requirements engineering processes (AI4RE), with a particular focus on requirements elicitation and automation of GUI prototyping such as automatic translation of user requirements into intuitive interface designs, and improving software engineering processes through intelligent tools (AI4SE).
Integrating World Foundation Models for Planning and Control in Robotics (Bachelor). This project investigates how World Foundation Models can be used to enhance robotic planning and control. The student will design and implement a system that leverages WFMs to understand environmental dynamics and generate action plans. Tasks may include multi-step planning, policy refinement via fine-tuning, and benchmarking in simulation environments (e.g., Isaac Sim). The objective is to develop an end-to-end, learning-augmented planning framework with improved adaptability and performance.
Exploring World Foundation Models for Robot Navigation (Master). This project focuses on using pre-trained World Foundation Models to support basic navigation tasks in mobile robotics. The student will work with simulated environments to explore how these models can assist in perception and simple decision-making. The emphasis will be on understanding the capabilities of WFMs and integrating them into a basic planning pipeline.
Library Learning for Autoformalization. Autoformalization aims to translate informal mathematical text into formal proofs, a key challenge in AI-assisted theorem proving [4,5]. This thesis investigates how existing library learning techniques, e.g. [1,3], impact the effectiveness and efficiency of autoformalization when integrated with Large Language Models and program synthesis. The goal is to evaluate their ability to reuse previously generated lemmas and improve proof success rates. The study will use benchmark datasets like miniF2F and MATH and compare against baselines without library learning. Results will inform the practical value and limitations of library learning [2] in formal reasoning systems.
Analyzing Exploration and Intrinsic Motivation of Large Language Models. This thesis aims to enhance the exploration capabilities of large language models, enabling them to intelligently navigate diverse environments. You will tackle and evaluate state-of-the-art agents in tasks such as web browsing, recommendation systems, and complex games. Hereby, we offer a broad range of directions for your thesis, including self-improvement and large model distillation.
Developing a Copilot for Building Information Modeling (BIM) Using AI-driven Assistance. Building Information Modeling (BIM) is an essential process in modern architecture, engineering, and construction (AEC), facilitating collaboration and efficiency throughout a project’s lifecycle. However, the increasing complexity of BIM models and workflows creates challenges in usability, error detection, and design optimization. Inspired by recent advancements in AI-assisted programming environments, this thesis explores the development of a BIM Copilot—an intelligent assistant that enhances the BIM workflow by providing real-time recommendations, automated error detection, and design optimizations. When a user begins designing a BIM model, the Copilot assists by suggesting the next logical BIM components based on the existing structure and design patterns. As the user defines a new component, the system can autocomplete its parameters by analyzing contextual elements such as material properties, dimensions, and spatial relationships. This ensures consistency and accelerates the modeling process, reducing manual input errors and enhancing workflow efficiency. The implementation is evaluated through user studies and empirical benchmarks to measure improvements in speed, accuracy, and overall user experience.