Task-Level Programming by Demonstration for Mobile Robotic Manipulators through Human Demonstrations based on Semantic Skill Recognition

Abstract:
The community has developed many concepts to facilitate the programming of industrial and service robots for non-expert users. Despite the increasing popularity of mobile robotic manipulators in various industries, most of the intuitive programming concepts are tailored to static manipulators. We developed a concept for novice user programming of mobile manipulators based on the Task-Level Programming by Demonstration (TLPbD) approach. To transfer the approach onto the mobile system, we integrate an interface for programming tasks through human demonstration by extracting the observed skills from the camera stream of the robotic system. Our approach allows the user to adapt and parameterize the created program after the demonstration as well as to integrate the demonstrated task into a sequence of tasks. We evaluated our approach in a user study with ten participants in comparison to a standard Task-Level Programming (TLP) method. With our approach, the users were on average four to five times faster in programming the specified task. In addition, we assessed the mental workload and the perceived usability and can show a significant advantage in comparison to the original TLP method.