Soft-Object Grasping Object

    The dataset contains two datasets and 36 python scripts.    dataset_1:Thirty‒five healthy adults (23 males and 12 females, aged 22 ± 3 years) were recruited for the experimental test. Their right hands were equipped with twenty‒two force sensing resistor (FSR) sensors (FSR 402, Interlink Electronics, Inc., USA) placed on positions according to the literature. Two sensors were on the thumb, twelve sensors were on the four fingers, respectively and eight sensors were on the palm. Two rubber airbags of similar geometries (cylindrical and spherical shapes) were chosen as soft objects for grasping. Additionally, we added an ellipsoid airbag with a shape to increase the difficulty of identification. An air pressure sensor (CJGR‒15, Xi’an CET Co., Inc., China) was connected to each airbag via silicone tubes to record the change in the internal pressure (IP). In the experiment, the participants were instructed to grasp the airbags with an initial IP of 10 kPa and maintain their grasping configuration or strength while the airbags were subjected to deformation of a certain degree (IP = 20 kPa, 30 kPa and 40 kPa). The participants repeated their grasping movements 10 times (10 trials) for each airbag with each degree of deformation. In accordance with the airbag IP data, we classified the grasping movement into three phases: loading, maintenance, and release . Consequently, the tactile data (sampling frequency = 50 Hz) were extracted from the 22 FSR sensors in the grasping‒maintenance phase to form a dataset (n = 2870, after the exclusion of 280 invalid trials from a total of 3150 trials), which was subsequently divided into a training set (70%, n = 2010) and a test set (30%, n = 860).     dataset_2: In accordance with the optimal placement, we mounted sensors on a polyester glove to enable accurate identification of soft objects. To examine the performance of the glove, we performed an experimental test on identifying soft objects during grasping. In this experiment, thirty‒five participants were instructed to grasp the three rubber airbags 10 times (10 trials) each while wearing the glove (Fig. 4b). In each trial, each participant performed a grasping action with arbitrary strength. We collected the tactile data acquired from the glove to construct a new dataset (n = 960) that was then divided into a training dataset (70%, n = 670) and a test set (30%, n = 290).     36 python scripts: we used six common ML classifiers, including logistic regression (LR), support vector machine (SVM), K‒nearest neighbors (KNN), linear discriminant analysis (LDA), random forest (RF) and multilayer perceptron (MLP), to examine the efficiency of the all original sensors (12 python scripts) the selected sensors on the hand (12 python scripts) and the glove (12 python scripts) in soft‒object identification.