Tactile Based Fabric Classification via Robotic Sliding

#artificialintelligence 

Tactile sensing endows the robots to perceive certain physical properties (which are not directly viable to visual and acoustic sensors) of the object in contact. Robots with tactile perception are able to identify different textures of the object touched. Interestingly, textures of fine micro-geometry beyond the nominal resolution of the tactile sensors, can also be identified through exploratory robotic movements like sliding and rubbing. To study the problem of fine texture classification via robotic sliding, we design a robotic sliding experiment using daily fabrics (as fabrics are likely to be the most common materials of fine textures). We propose a feature extraction process to encode the acquired tactile signals (in the form of time series) into a low dimensional (<= 7D) feature vector. The vector captures the frequency signature of a fabric texture such that distinctive fabrics can be classified by their correspondent feature vectors. The experiment includes multiple combinations of sliding parameters, i.e., speed and pressure, for the investigation into the correlation between sliding parameters and the generated feature space. Results show that changing the contact pressure can greatly affect the significance of the extracted feature vectors. For our specific sensor used in the experiments, there exists a sweet spot of pressure for the fabric classification task. Adversely, variation of sliding speed shows no apparent impact on the performance of the feature ext...