Making Sense of Touch: Unsupervised Shapelet Learning in Bag-of-words Sense
Zhicong Xian, Tabish Chaudhary, Jürgen Bock
TL;DR
This work tackles unsupervised learning of discriminative time-series shapelets for tactile robotic data by introducing NN-STNE, a neural architecture that embeds shapelet-based representations through a t-SNE hidden layer. The model processes time-series via sliding windows and NCC-based similarity, then maps distances to shapelets into a low-dimensional, interpretable feature space used for clustering, while preserving local structure with a Gaussian-kernel spectral objective and promoting diversity with a shapelet-weight regularization scheme. Evaluations on UCR subsets and a robot switch-on task show that NN-STNE improves clustering performance (on average about 16.7% over strong baselines like UDFS) when used as a pre-clustering feature learner, highlighting the practical value of non-linear, unsupervised shapelet learning for interpreting touch data. The approach offers a scalable, interpretable pathway for enhancing recognition and diagnosis in contact-rich robotic applications, bridging time-series shape features with unsupervised, high-level clustering insights.
Abstract
This paper introduces NN-STNE, a neural network using t-distributed stochastic neighbor embedding (t-SNE) as a hidden layer to reduce input dimensions by mapping long time-series data into shapelet membership probabilities. A Gaussian kernel-based mean square error preserves local data structure, while K-means initializes shapelet candidates due to the non-convex optimization challenge. Unlike existing methods, our approach uses t-SNE to address crowding in low-dimensional space and applies L1-norm regularization to optimize shapelet length. Evaluations on the UCR dataset and an electrical component manipulation task, like switching on, demonstrate improved clustering accuracy over state-of-the-art feature-learning methods in robotics.