In the circular factory, the reprocessing of used products into the newest product generation requires complex dis-/ and reassembly manipulation actions. Additionally, this involves dealing with uncertain product specifications such as rust or contamination, as well as unknown conditions such as damaged or previously disassembled products. The central question addressed by the working group is how to learn from observing humans performing these complex manipulation tasks and how to transfer this knowledge to automated execution by robots.

A fundamental element of the working group is the learning cell. In this experimental setup, humans are captured using multiple modalities, such as RGB-D video cameras and or eye tracking. Subsequently, robot-processable task models are developed from this data, with the objective to reproduce and generalize the learned actions on a bimanual robotic arm setup.

Leading Principal Investigator