Came across a nice paper by OpenAI's Ian Goodfellow and Google Brain on using video prediction to model a robot's "after-effects":

Unsupervised Learning for Physical Interaction through Video Prediction

http://arxiv.org/abs/1605.07157