Webseite Werkzeugmaschinenlabor (WZL) der RWTH Aachen


Machining of large components (e.g. for the aviation industry) is almost exclusively conducted on machine tools (aka CNC machines). Yet, the workspace of these machines is small compared to their size which significantly raises the cost of large components.

Industrial robots (IR) denote cost as well as space efficient alternatives to conventional machining tools. Unfortunately, IR are usually less rigid, which negatively impacts the workpiece quality.

The effects of the low rigidity on work-piece quality can be mitigated by using model-based feedforward control. Naturally, such a control approach relies on an accurate model of the system’s dynamics. Yet, obtaining a sufficiently accurate dynamics model constitutes a considerable challenge.

Problem statement

The two common approaches for obtaining a dynamics model are analytical and data-driven modeling. Analytical models are based on physics equations such as the Newton-Euler equations of motion. Analytical models are human interpretable and can potentially extrapolate to unseen regions of the state-space. However, many phenomena in robotic systems such as friction are difficult to model a priori causing considerable model errors.

Data-driven models such as neural networks require less prior knowledge by fitting a function directly to the in- and output data of the system. However, this increase in flexibility requires considerable amounts of data and the predictions of such models do not extrapolate well.

Structured models aim to combine the advantages of these approaches by incorporating available structural knowledge (e.g. in form of analytical functions) into data-driven models. The resulting model predictions are supposed to comply with system constraints while improving generalization and data efficiency.

The goal of this thesis is to obtain a structured model of a robot arm’s dynamics by combining neural networks with an analytical dynamics model.

Your tasks

  • Generation and preparation of training data on a robot arm using standard control approaches
  • Research on combining neural networks with robot modeling
  • Implementation and training of a structured robot dynamics model

Your profile

  • High motivation and interest in solving mathematical problems in robotic systems
  • Solid programming experience
  • Good English or German skills

What we offer:

  • Industrial robot machining testbed
  • Extensive supervision by WZL as well as DSME / Max Planck Institute for Intelligent Systems, Stuttgart


  • CV, grade transcript and other relevant documents (optional)



Minh Trinh, M.Sc. RWTH
Steinbachstr. 25
D-52074 Aachen
Telephone: +49241 / 80 27457

Um sich für diesen Job zu bewerben, sende deine Unterlagen per E-Mail an m.trinh@wzl.rwth-aachen.de