ID 2446: EmpkinS C01: Physics-Informed Deep Learning for Movement Analysis and Control

Bachelor’s / Master’s Thesis / Research Internship / Project

The real-time control and analysis of human poses and movements has long been a complex problem in fields like computer vision and machine learning. Physics-informed neural networks (PINNs) have recently shown promise in improving the accuracy of pose estimation, but there is still a lot of potential to explore.

If you’re interested in investigating the applications of physics-informed learning for difficult-to-solve tasks, a number of student projects are available. We can work together to tailor a project that aligns with your skills and interests. Projects can start immediately, and thesis work can begin from April 1st.

Feel free to reach out if you’re interested in learning more or discussing potential topics.

Example Tasks

• Explore how physics informed machine learning can solve predictive control tasks
• Explore the interactions between physics-informed learning and PD controllers
• Parameter identification and transfer learning
• Implementation of multibody dynamics models
• Depending on the topic and scope of your project, you will:
  • Implement physical and neural models in sympy and PyTorch
  • Work with pre-existing datasets or solve unsupervised tasks
  • Document your code in a clear and structured manner

Requirements (depending on the topic and type of project)

• Interest in diving into the fields of movement analysis, robotics and optimization
• Knowledge of deep learning, experience with PyTorch or Tensorflow (or willingness to learn such!)
• Further skills:
  • Knowledge in data processing, data analysis, and data visualization using Python.
  • Basic knowledge of mechanics is a plus!

Supervisors

Markus Gambietz, M. Sc.

Researcher & PhD Candidate