Robots that can manipulate objects in unstructured environments and\ncollaborate with humans can benefit immensely by understanding natural\nlanguage. We propose a pipelined architecture of two stages to perform spatial\nreasoning on the text input. All the objects in the scene are first localized,\nand then the instruction for the robot in natural language and the localized\nco-ordinates are mapped to the start and end co-ordinates corresponding to the\nlocations where the robot must pick up and place the object respectively. We\nshow that representing the localized objects by quantizing their positions to a\nbinary grid is preferable to representing them as a list of 2D co-ordinates. We\nalso show that attention improves generalization and can overcome biases in the\ndataset. The proposed method is used to pick-and-place playing cards using a\nrobot arm.\n