Towards a computational model of spatial perspective taking