This demo is a part of the final deliveries of CloPeMa. In this piece of work, we adapted our stereo vision system to autonomous laundry tasks, we proposed a geometry-based feature architecture to parse the configurations of the deformable clothing. In order to have a fully understanding, we build a rich visual representation with hierarchical structures: from lower-level curvatures features, to mid-level shape and topologies features, finally get high-level semantic wrinkle representation. This feature architecture can be used for multiple laundry task, such as grasping, flattening and now has been extended to sorting. Our experimental results show that, our depth sensor outperforms kinect-like cameras in sensoring garments.
Li Sun, et. al. “Accurate Garment Manipulation using Binocular Stereo Head with the Application to Dual-Arm Flattening”, ICRA 2015. This paper got the best conference paper nomination.
In this research, we propose a robot vision pipeline to recognise clothing categories from 2.5D features. Two novel features, BSP (B-Spline Patch) and TSD (Topology Spatial Distances) are devised for this task, together with an improved Locality-constrained Linear Coding technique. This is the first research work to integrate clothing category recognition where clothes are in unconstrained and random configurations into a robot autonomous sorting pipeline. To verify the performance of our proposed method, we create a dataset of 50 clothing items of 5 categories sampled in random configurations (a total of 2100 clothing samples). Experimental results show that our approach is able to reach 83.2% accuracy while classifying unseen clothing items.
Li Sun, Gerarado Aragon-Camarasa, Simon Rogers, J. Paul Siebert. "Single-Shot Clothing Category Recognition in Free-Configurations with Application to Autonomous Clothes Sorting", IROS 2017.
In this research, we propose a Gaussian-Process-based interactive perception approach for recognising highly wrinkled clothes. We have integrated this recognition method within a clothes sorting pipeline for the pre-washing stage of an autonomous laundering process. Our approach differs from reported clothing manipulation approaches by allowing the robot to update its perception confidence via numerous interactions with the garments. The classifiers predominantly reported in clothing perception (e.g. SVM, Random Forest)studies do not provide true classification probabilities, due to their inherent structure. In contrast, probabilistic classifiers (of which the Gaussian Process is a popular example) are able to provide predictive probabilities. In our approach, we employ a multi-class Gaussian Process classification using the Laplace approximation for posterior inference and optimising hyper-parameters via marginal likelihood maximisation.
Li Sun, Simon Rogers, J. Paul Siebert. "Recognising the Clothing Categories from Free-Configuration using Gaussian-Process-Based Interactive Perception", ICRA 2016.