Relational reinforcement learning and recurrent neural network with state classification to solve joint attention | IEEE Conference Publication | IEEE Xplore