This master thesis explores the integration of physical laws and constraints into graph-based geometric reconstruction processes. The research focuses on developing optimization frameworks that not only reconstruct geometric structures represented as graphs but also ensure the resulting structures adhere to fundamental physical principles. The aim is to achieve a more stable and reliable optimization. By incorporating physics-based constraints such structural stability and physical feasibility, the reconstruction process produces results that are both geometrically accurate and physically plausible.
You will develop a mathematical framework for physics-constrained graph optimization that incorporates relevant physical properties into the reconstruction process. The framework will be evaluated on a FEM case studies.
- Knowledge of mathematical machine learning fundamentals.
- Proficiency in at least one graph processing framework (PyG, DGL, GraphScope).
- Experience with neural network architectures.
- Programming in Python, C++ or Zig.
Sebastian Baum
♦
The trend toward the autonomous operation of industrial plants requires new approaches to inspection and maintenance processes. While the operation of many systems is already largely automated, maintenance and inspection tasks are still predominantly carried out manually by skilled personnel and contribute significantly to overall life-cycle costs. In addition to high labor expenses, human intervention introduces safety risks and potential sources of error, particularly in hard-to-reach or hazardous environments.
A particularly relevant application area is the autonomous operation of offshore platforms, for example for the production of power-to-X products from wind energy, as in the H2Mare project. Large distances, harsh environmental conditions, and limited transportation options make personnel deployment extremely challenging. In such contexts, legged robots—such as humanoids or robotic dogs—can make a decisive contribution. Owing to their mobility and adaptability, they can navigate complex, human-designed environments, operate valves, inspect sensors, or detect leaks. In doing so, they can relieve maintenance personnel, reduce safety risks, and minimize downtime through faster responses to malfunctions.
The objective of this work is to investigate and conceptually develop application scenarios for legged robots in industrial environments, particularly in the field of inspection and maintenance tasks.
As part of a literature review, the technical fundamentals of legged robots in industrial contexts will first be examined. This includes identifying the sensory, motor, and cognitive capabilities of these systems, as well as analyzing the current state of the art. Subsequently, different types of robots and their potential industrial applications will be classified.
In parallel, an analysis of task profiles and requirements for inspection and maintenance activities performed by human personnel will be conducted. This analysis serves as the basis for comparing the capabilities of legged robots with the requirements of real-world inspection and maintenance tasks in order to evaluate their feasibility and suitability for potential application scenarios. The evaluation and analysis of task profiles will be carried out using scientific methods such as Hierarchical Task Analysis. For the identification of task profiles and requirements, the classifications provided by the German Federal Employment Agency and the ESCO framework of the European Commission will be used as a starting point.
Building on these findings, feasible application scenarios will be identified, and a concept for their technical implementation will be developed. The practical implementation of the application cases will be examined using the Unitree Go1 robotic dog and the Unitree G1 humanoid robot. In addition, the necessary software stack for realizing these applications will be designed.
As part of the implementation phase, a 3D simulation environment will be established. For this purpose, assets will be created in a 3D modeling program (e.g., Blender or FreeCAD) to enable a simulated investigation of the developed application scenarios. The 3D models should be compatible with common robotic simulation environments such as Gazebo, Isaac Sim, or MuJoCo. Optionally, the integration of the 3D models and the existing digital models of the Unitree Go1 and G1 robots into one of these simulation environments will be carried out.
- Good command of English (a large portion of the literature for the review will be in English).
- Experience in the field of robotics (ROS) and familiarity with 3D modeling software (e.g., Blender) are advantageous.
- High degree of independence, initiative, and motivation.
Peter Frank
♦