Grasping States Classification Based on Wavelet Fuzzy Entropy and t-SNE

The classification of grasping states is crucial for robots' stable operations that based on its real-timely results, the robot can adjust the appropriate gripping force. However, until now, classifying stable grasping states of multiple objects in different scenarios is still a challenging issue since it needs to find the commonality of different objects in the same state. Aiming to solve this problem, we present an effective stable grasping states classification method based on the combination of wavelet fuzzy entropy and t-Distributed Stochastic Neighbor Embedding (t-SNE), and deploys it to the robot platform to complete the state feedback. Firstly, the wavelet decomposition of the tactile sensor signal is performed and the fuzzy entropy of the signal is calculated, and the time-domain feature variables are extracted together to form the eigenvector set. Secondly, t-SNE is used to reduce the dimensionality and visualize the high-dimensional data. Finally, the extracted features are input into Support Vector Machine (SVM) to classify the stable state of the t-SNE data after dimensionality reduction. The experimental results show that the proposed method has strong stability recognition ability, the accuracy rate reaches 90.37%, the average success rate of grasping state feedback reaches 91.1%, and the average success rate is increased by 5.53% compared with the Principal Component Analysis (PCA) feature processing method.