A web-based infrastructure for recording user demonstrations of mobile manipulation tasks

摘要: Learning from demonstration (LfD) is a common technique applied to many problems in robotics, such as populating grasp databases, training for reinforcement learning of high-level skill sets and bootstrapping motion planners. While approaches are generally highly valued, they rely on the often time-consuming process gathering user demonstrations, hence it becomes difficult attain sizeable dataset. In this paper, we present tool capable recording large numbers high-dimensional demonstrations mobile manipulation tasks provided by non-experts field. Our accomplishes via web interface that requires no additional software be installed beyond browser, well scalable architecture supporting 10 concurrent demonstrators single server. employs lightweight simulation environment reduce unnecessary computations improve performance. Furthermore, show how our can used gather set task leveraging existing crowdsource platforms. The data collected has been made available robotics community. We also experiments which apply through infrastructure teach robot grasp, perform dexterous scooping accelerate planning full-body tasks.