Discrimination method for robot stable grasping based on tactile array perception

To address the issue that traditional single-point pressure sensors failed to provide a comprehensive reflection of the robot's grasping contact state，making it difficult to achieve accurate grasping status discrimination，an efficient and highly accurate robot stable grasping discrimination method was proposed based on the multi-point sensing characteristics of tactile array sensors．First，by capturing the distributed force information when the robot interacted with different objects during grasping，a mapping between the distributed force and tactile images was established，thereby constructing a dataset of robot grasping tactile images．Using a multi-layer perceptron framework，a grasping status discrimination model was developed to classify the robot's grasping states． Then，through training and optimizing the multi-layer perceptron model with different numbers of layers and nodes，the optimal parameters for the grasping status discrimination model were obtained． Furthermore，the proposed method was compared with various learning-based grasping discrimination algorithms． Results show that the proposed grasping state discrimination method has a discrimination accuracy of 99.74% and an average time of 2.3 ms，surpassing the baseline algorithms in both discrimination accuracy and speed． Through real-world grasping experiments，the method achieves a discrimination accuracy of 94%，which fully proves its strong robustness in stable grasping discrimination for objects outside of the dataset. © 2024 Huazhong University of Science and Technology. All rights reserved.