Visual Task-Level Programming by Demonstration for Mobile Robotic Manipulators

Abstract

Robot programming concepts for non-expert users are crucial for deploying robotic applications in a broad range of industries and domains. Especially in the field of mobile robotic manipulators, we see a lack of methods and approaches for enabling novice users to adapt the robot program according to their needs. Within this thesis, we develop a concept for the intuitive programming of a mobile manipulator and prove its usability in a user study. As a starting point serves the Task-Level Programming by Demonstration (TLPbD) approach, which was developed in 2019 by Steinmetz et al. on a static manipulator. It combines two of the most common approaches in the field: Task-Level Programming (TLP) and Programming by Demonstration (PbD). We take into account new findings in TLP and PbD and compare the differences in Human-Robot-Interaction (HRI) among mobile and static manipulators with a focus on laboratory automation. Our core contribution is the change of the demonstration modality from kinesthetic teaching to a visual demonstration through the user. This allows the user to create a task-level program by demonstrating the task and still (re-)parameterize or adapt the program afterwards, for example for integrating it in a sequence of multiple tasks. The concept is implemented into the existing TLP framework RAZER and integrated on a mobile robotic manipulator. The user study is conducted with employees of the German Aerospace Center (DLR). We assess the programming time, the mental workload of the users as well as the usability of our approach in comparison to a standard TLP approach. The results show a clear reduction in programming time using our method. Also, our approach achieves a higher usability and a reduced mental workload in comparison to the common TLP method.