Transferring Human Manipulation Knowledge to Robots with Inverse Reinforcement Learning

Abstract

The need for adaptable models, e.g. reinforcement learning, have in recent years been more present within the industry. In this paper, we show how two versions of inverse reinforcement learning can be used to transfer task knowledge from a human expert to a robot in a dynamic environment. Moreover, a second method called Principal Component Analysis weighting is presented and discussed. The method shows potential in the use case but requires some more research.