A household robot is expected to perform a wide variety of manipulating operations with an understanding of the purpose of the task. To this end, a desirable robotic application should provide an on-site robot teaching framework by non-experts. Here we propose to apply the Learning-from-Observation (LfO) framework to grasp-manipulation-release class household operations. The framework maps a human demonstration to predefined task models through a one-shot demonstration. A task model contains both high-level knowledge about the geometric constraints of tasks and low-level knowledge about human postures. The key idea is to design a task model that 1) covers a wide variety of household operations and 2) contains human postures to achieve tasks. We verified the effectiveness of our framework by testing an implemented system with several operations. In addition, we quantified the coverage of the task model by analyzing Web videos about household operations. In the context of one-shot robot teaching, the contributions of the paper are: 1) to propose a framework that 1) covers various tasks in grasp-manipulation-release class household operations and 2) mimics human postures during the operations.
View on arXiv