A Model of Generating and Predicting Intention toward Human-Robot Cooperation

摘要

We normally tend to think that we, as human beings, act as if we have intention to achieve a goal. However, in the situation of cooperating with other people, it is difficult to come up with a mechanism for predicting and adjusting each other's intention when we think that the goal is given top-down. Thinking that intention comes from context may help to solve that problem. In this study, we attempt to model an action learning mechanism that generates/predicts self-intention using deep Q-network and recurrent neural network. And then we show such a mechanism makes it possible for the agent to predict intention of others and goals.