Machine learning

We search for the learning effect!

The goal of the LEROSH approach is an intelligent and interacting system, which can process components automatically – taught by the craftsman. Critical here is the formalization of the expert’s knowledge of how components are to be processed, the simplification of operation so that no programming knowledge is required, and the automation of processing for batch size 1.

This is only possible by making robots capable of learning, i.e. by allowing robots to learn how to process certain components, forms and materials on the basis of demonstrations and the skills of the craftsmen. Contact forces and other sensor data should be increasingly interpreted for this purpose and evaluated with machine learning to improve execution in terms of quality and possible component complexity. In addition, this information from the measurements can be used for automatic recognition of machining classes (e.g. geometry, material or quality), thus enabling extrapolation of the learned skills to unknown components.