Imitation learning is a powerful paradigm for robot skill acquisition.However, obtaining demonstrations suitable for learning a policy that maps fromraw pixels to actions can be challenging. In this paper we describe howconsumer-grade Virtual Reality headsets and hand tracking hardware can be usedto naturally teleoperate robots to perform complex tasks. We also describe howimitation learning can learn deep neural network policies (mapping from pixelsto actions) that can acquire the demonstrated skills. Our experiments showcasethe effectiveness of our approach for learning visuomotor skills.
展开▼