Magnetostrictive-based multimodal tactile sensors for object recognition

Kaile Liu,Ling Weng,Boyang Hu,Zhuolin Li,Yang Liu,Yuxin Chen,Shengwang Jiang
DOI: https://doi.org/10.1063/5.0178609
IF: 1.697
2024-01-01
AIP Advances
Abstract:Tactile perception of the human hand plays a critical role in everyday object recognition. The development of multimodal tactile sensors that can sense stimuli with high sensitivity and low cost is important for intelligent perception. In this article, a multimodal tactile sensor attached on a mechanical hand is studied, which consists of a magnetostrictive tactile sensor, a temperature sensor, and a flex sensor. By applying multimodal tactile sensors to a robotic hand to grasp objects, the output voltage of the magnetotactictive tractile tactile sensor can be used for object shape and softness recognition. The bending angle of the knuckles can be obtained by the flex sensor for object contour size recognition, while the temperature distribution of objects can be obtained through temperature sensors. In order to improve the accuracy, a 1-D convolutional neural network-extreme learning machine (CNN-ELM) pattern recognition model based on the combination of 1-D CNN and ELM is presented, with the accuracy of 97.14% for 21 objects. This multimodal tactile sensor has promising applications in the field of tactile intelligence and humanoid robotics.
materials science, multidisciplinary,physics, applied,nanoscience & nanotechnology
What problem does this paper attempt to address?
The paper attempts to address the problem of implementing multimodal tactile sensors in robotic hands to improve the accuracy and efficiency of object recognition. Specifically, the research aims to develop a multimodal tactile sensor capable of perceiving the shape, hardness, temperature, and contour size of objects, and apply it to a robotic hand. Through machine learning methods (such as the 1-D Convolutional Neural Network-Extreme Learning Machine (CNN-ELM) model), the goal is to achieve high-precision recognition of 21 different objects. The main contributions of the paper include: 1. **Designed a tactile sensor based on the magnetostrictive effect**, used to perceive the shape and hardness of objects, characterized by high sensitivity and low cost. 2. **Integrated a bending sensor**, used to measure the bending angle of finger joints, thereby recognizing the contour size of objects. 3. **Incorporated a temperature sensor**, used to detect the temperature distribution of objects, further enriching the tactile information. 4. **Proposed a 1-D CNN-ELM pattern recognition model**, combining the advantages of one-dimensional convolutional neural networks and extreme learning machines, achieving high-precision classification and recognition of various objects with an accuracy rate of 97.14%. The combination of these technologies allows the robotic hand to obtain more comprehensive information when grasping objects, improving the accuracy and reliability of object recognition. This has broad application prospects, especially in the fields of tactile intelligence and humanoid robots.