Convolutional Encoder-Decoder Networks for Robust Image-to-Motion Prediction.