MuPNet: Multi-modal Predictive Coding Network for Place Recognition by Unsupervised Learning of Joint Visuo-Tactile Latent Representations.