Robot screwing state detection based on tactile sequences

The task of screwing in a robot assembly is consid-ered challenging, and currently, there is no generalized solution available to determine the tightness of the connection between a bolt and nut. In this paper, we propose a method to classify the screwing state by utilizing time-sequential contact measurements. Our approach involves incorporating tactile information obtained from tactile sensors, designing a module for processing time-sequential tactile features, introducing an attention mechanism, establishing a network for detecting the screwing state in robots, and constructing a dataset specifically for the robot screwing state. We train a convolutional neural network on this dataset to classify the screwing state based on the tactile information. Ultimately, we design a visual and tactile robotic screwing control system to autonomously perform screwing operations. Simulation and experimental results demonstrate the efficacy of our method for autonomous screwing assembly tasks in uncertain conditions.