Visual Imitation with Reinforcement Learning using Recurrent Siamese Networks