Active intention inference for robot-human collaboration

Abstract

Understanding human intention is an important ability for an intelligent robot to collaborate with a human to accomplish various tasks. During collaboration, a robot with such ability can predict the successive actions that a human partner intends to perform, provide necessary assistance and support, and remind for the missing and failure actions from the human to achieve a desired task purpose. This paper presents a framework that allows a robot to automatically recognize and infer the action intention of a human partner based on visualization, in which an inverse-reinforcement learning (IRL) system is learnt based on the observed human demonstration and used to infer the human successive actions. Compared to other systems based on reinforcement learning, the reward of a Markov-Decision process (MDP) is directly learned from the demonstration. In our experiment, we provide some examples of the proposed framework which yields promising results with coffee-making and pick-and-place tasks. Regarding to the human-intention model based on IRL, the coffee-making experiment indicates that the action is globally predicted because the action of putting down the water pot is selected instead of pouring water when the cup is already filled with water.