Spectro-Temporal Recurrent Neural Network for Robotic Slip Detection with Piezoelectric Tactile Sensor

摘要

In this paper, we present a novel tactile-based method for detecting slippage in robotic manipulation, using a single piezoelectric sensor. The method combines spectral analysis (FFT) and deep learning (GRU) for improved efficiency and adaptability. We implement an automated data-collection process with accurate and unbiased labels of slip events. The proposed method is evaluated through an ablation study characterizing the influence of model hyperparameters and interaction settings. The results show a high classification accuracy of 98.70% at 100Hz and detection delays of 8.5 ± 23.7ms, demonstrating the relevance of our spectro-temporal pipeline. The proposed method has the potential to enhance the performance of robotic systems and increase their reliability in robotic grasping applications.