Industrial robots are an integral part of modern production systems. Their flexibility and comparatively low costs are impressive, but they have significantly lower absolute and path accuracy than machine tools. For many high-precision machining tasks - such as drilling, milling or joining - the accuracy of commercially available robots is not sufficient. This is why they have so far been used primarily for handling, welding or simple assembly tasks. In addition, robot systems have to meet specific functionality and safety requirements before they can be put into operation.

The Institute for Control Engineering of Machine Tools and Manufacturing Units (ISW) at the University of Stuttgart is researching solutions to close these gaps. The aim is to make robots usable for high-precision manufacturing processes and to virtually validate their systems before commissioning.

The ISW approach combines control engineering methods with physical modeling and machine learning methods. Virtual commissioning (VIBN) enables comprehensive system tests in progressive development stages.

From the classic model to AI support

Traditional methods for increasing accuracy rely on kinematic calibration. Geometric deviations are measured and corrected. However, this does not cover all sources of error. In particular, the transmission errors of the gears in the robot axes, which are caused by assembly and manufacturing tolerances, have not yet been taken into account. These have a direct effect on the end effector position and accuracy.

While such effects can be described classically using mathematical functions such as Fourier series, the ISW also pursues a data-driven approach: machine learning algorithms are able to recognize complex, non-linear dependencies from measurement data.

AI methods for the modeling of transmission errors

The development of an ML model begins with systematic data acquisition. This involves recording the drive and output-side angles of the robot axes. Regression models are then trained, which capture non-linear relationships and map a large number of error sources simultaneously. In contrast to Fourier series models, which are based on a fixed mathematical structure, data-driven models adapt flexibly to the robot in question.

Demonstrator at the trade fair

At the trade fair, ISW will be exhibiting a Stäubli "TX2-40" with a digital twin. Visitors can give the robot virtual points and transfer them into real programs. At the same time, the improved path accuracy through model-based error compensation will be demonstrated - a practical bridge from research to industrial application.

Potential for the production of tomorrow

The use of simulation, modeling and compensation of transmission errors opens up far-reaching perspectives for production engineering:

• In future, robots will take over tasks that were previously the preserve of machine tools.
• High precision with maximum flexibility: robotics and machining are growing closer together.
• Extensive system tests before commissioning.

The ISW thus contributes to the further development of industrial robots from pure handling devices to high-precision tools for flexible production.

The ISW at SPS 2025: Hall 6 Stand 340