Performative Body Mapping is a method for harnessing the embodied expertise of dancers to inform the design, movement and behaviour of non-anthropomorphic social robots. The method simplifies the correspondence problem through the novel use of costumes that allow much of the difficult human-robot mapping to be delegated to dancers. A mixture density recurrent neural network has been used to model sequences captured during movement studies to create new dance sequences in the style of the dancers inhabiting a costume.