AI-Supported Robot Trajectory Optimization

This experiment has been carried on by Kautenburger GmbH. Robot processing cells are constantly gaining relevance in production. Despite their low stiffness, they are also increasingly used for CNC machining. Kautenburger GmbH wants robots for the post-processing of cast metal and for white plaster products.The main problems come from machining inaccuracies due to the limited stiffness of the robots. In order to correct these errors at runtime, complex and accurate stiffness models and processing force models should be built as a general solution. However, the stiffness model parameters can only be acquired experimentally and since this solution cannot be transferred to other robots, this will become one of the goals of the experiment.

The experiment aimed to develop a possible optimization strategy with the help of AI methods, which then makes it possible to maintain the robot desired trajectory during machining tasks even without absolute measurement of the tool centre point (TCP) at runtime.

In the context of this project it was examined whether the errors can be mapped with the help of a machine learning approach. For this purpose, a robot with a processing spindle is absolutely measured at runtime. The trajectory thus determined is then compared with the TARGET trajectory, the TARGET 3D model and the 3D scan of the ACTUAL model. Based on the resulting deviations, a possible optimization is then derived with the help of AI methods, which then makes it possible to maintain the desired trajectory even without absolute measurement of the TCP at runtime.