Interaction identification through tactile sensing during cloth manipulation using a 3-axis touch sensor
Document typeConference report
Rights accessOpen Access
European Commission's projectCLOTHILDE - CLOTH manIpulation Learning from DEmonstrations (EC-H2020-741930)
Tactile feedback during cloth manipulation could be crucial in addressing the huge challenges involved in closing the loop during execution, complementing vision. However, up to our knowledge, tactile sensing has only been successfully used in cloth manipulation to classify type of fabrics, detect how many layers were grasped, and estimate the grasping force. In this work, we want to explore its potential to also provide information about whether the task is executed as expected. Two types of experiments are performed, in which a robot carries out 4 simple tasks, all involving a single finger manipulating a flat cloth on a table. Firstly, we analyze the sensor’s signals once the cloth manipulation has finished using Dynamic Time Warping (DTW) to see if they are informative enough to classify the tasks. Our results show that tactile feedback depends highly on the type of manipulated fabric. Secondly, we analyze the tactile feedback during the manipulation of the cloth using a recurrent neural network (RNN). For each sensor measurement, the RNN recognizes if the finger slides over the cloth, pulls it, flattens a fold in it, or if it’s about to lose contact, with 95.7% accuracy. These are promising results that show how tactile sensing has the potential of providing crucial information that would be very difficult to obtain with vision only.
CitationGeer, I. [et al.]. Interaction identification through tactile sensing during cloth manipulation using a 3-axis touch sensor. A: IROS Workshop on Robotic Manipulation of Deformable Objects (ROMADO). "The proceedings of the 2020 IROS Workshop on Robotic Manipulation of Deformable Objects (ROMADO)". 2020, p. 1-8.