Mandatory Internship / Thesis: Spatial Reasoning for Partially Observed Industrial Scenes
Summer Semester 2027 – Fixed Term of 3-6 Months
How can intelligent systems reason about spatial structure when relevant parts of their environment cannot be directly observed? In this project, you will investigate methods for spatial reasoning in industrial visual perception. Relevant settings include scenes affected by occlusions, limited viewpoints, missing observations, or incomplete sensor coverage. The goal is to study how learning-based systems can represent and reason about objects, spatial relationships, and constraints when the available visual information is incomplete.
The project is open with respect to methodology. Depending on the selected problem and scope, possible approaches include generative models, vision-language models, object-centric or structured representations, probabilistic methods, world models, neurosymbolic approaches, or combinations thereof.
Your Tasks
- You will investigate approaches for reasoning over incomplete spatial observations.
- You will compare selected representations and inference strategies, which may include structured, recursive, sequential, or uncertainty-guided approaches.
- You will select, adapt, or develop a suitable approach and implement it in a proof-of-concept prototype.
- You will evaluate the prototype on a selected task.
The expected outcome is a critical feasibility assessment for real-world industrial settings. The project should identify a suitable representation and inference strategy for the selected scenario, analyze its assumptions and limitations, and evaluate its potential through a proof-of-concept prototype. Well-founded negative results are equally valuable.
Your Profile
- Currently pursuing an advanced master's degree in Computer Science, Artificial Intelligence, Data Science, Mathematics, Robotics, Physics, Engineering, or a related field.
- Good foundation in machine learning, diffusion models, Python, and computer vision.
- Familiarity with PyTorch, generative models, spatial reasoning, object detection, or uncertainty estimation is an advantage.
- Interest in reading research literature, developing hypotheses, designing controlled experiments, and critically assessing whether a new method is suitable for an industrial problem.
- Ability to quickly grasp complex technical problems, work independently and systematically, and communicate findings clearly within an interdisciplinary team.
- Strong interest in generative AI, visual reasoning, uncertainty-aware inference, and exploratory industrial research.
- Good written and spoken English skills.
You don't meet every requirement yet? We still encourage you to apply. We value curiosity, a strong willingness to learn, and the ability to approach complex problems systematically. You can also develop relevant skills during your thesis.
Sectors
Sectors
Summer Semester 2027 – Fixed Term of 3-6 Months
How can intelligent systems reason about spatial structure when relevant parts of their environment cannot be directly observed? In this project, you will investigate methods for spatial reasoning in industrial visual perception. Relevant settings include scenes affected by occlusions, limited viewpoints, missing observations, or incomplete sensor coverage. The goal is to study how learning-based systems can represent and reason about objects, spatial relationships, and constraints when the available visual information is incomplete.
The project is open with respect to methodology. Depending on the selected problem and scope, possible approaches include generative models, vision-language models, object-centric or structured representations, probabilistic methods, world models, neurosymbolic approaches, or combinations thereof.
Your Tasks
- You will investigate approaches for reasoning over incomplete spatial observations.
- You will compare selected representations and inference strategies, which may include structured, recursive, sequential, or uncertainty-guided approaches.
- You will select, adapt, or develop a suitable approach and implement it in a proof-of-concept prototype.
- You will evaluate the prototype on a selected task.
The expected outcome is a critical feasibility assessment for real-world industrial settings. The project should identify a suitable representation and inference strategy for the selected scenario, analyze its assumptions and limitations, and evaluate its potential through a proof-of-concept prototype. Well-founded negative results are equally valuable.
Your Profile
- Currently pursuing an advanced master's degree in Computer Science, Artificial Intelligence, Data Science, Mathematics, Robotics, Physics, Engineering, or a related field.
- Good foundation in machine learning, diffusion models, Python, and computer vision.
- Familiarity with PyTorch, generative models, spatial reasoning, object detection, or uncertainty estimation is an advantage.
- Interest in reading research literature, developing hypotheses, designing controlled experiments, and critically assessing whether a new method is suitable for an industrial problem.
- Ability to quickly grasp complex technical problems, work independently and systematically, and communicate findings clearly within an interdisciplinary team.
- Strong interest in generative AI, visual reasoning, uncertainty-aware inference, and exploratory industrial research.
- Good written and spoken English skills.
You don't meet every requirement yet? We still encourage you to apply. We value curiosity, a strong willingness to learn, and the ability to approach complex problems systematically. You can also develop relevant skills during your thesis.
