A High-Spatial-Resolution Magnetorheological Elastomer Tactile Sensor for Texture Recognition

With the popularization of robots in various fields such as industry and healthcare, tactile perception ability has gradually become the key to achieving precise operation of physical objects by robots. Although existing tactile sensors can enable robots to accurately detect pressure, shear force, and strain, the ability of robots to perceive the environment, classify objects, and recognize textures is equally important. To recognize fine surface textures, this article designs a high-spatial-resolution tactile sensor based on magnetorheological elastomers (MREs), with a texture recognition accuracy of 0.1 mm. Additionally, a texture recognition model combining the Kolmogorov-Arnold Network (KAN) and the bidirectional long short-term memory (Bi-LSTM) network is proposed, enhancing the model's ability to learn nonlinear relationships and thus improving texture recognition accuracy. An online training strategy is also introduced for this model, endowing it with certain generalization capabilities. Experimental results demonstrate that the tactile sensor achieves a recognition accuracy of 100% for 54 textures under fixed contact conditions, 96.23% accuracy under variable contact conditions, and 98% accuracy for new textures outside the dataset. This tactile sensor not only enables robotic object perception but also has potential applications in virtual reality and human-computer interaction.