講演情報
[8p-N201-10]Trial for Real-Time Object Recognition with CNT-PDMS Haptic Pressure Sensor
〇Ahmet KARACALI1, Kouki KIMIZUKA1, Yuki USAMI1,2, Hirofumi TANAKA1,2 (1.LSSE, Kyutech, 2.Neumorph Center)
キーワード：
Haptic Sensor、Real-Time Processing、Material AI
Introduction: With the advancement of technology, artificial devices and sensors that can mimic human touch and interact with the environment have become an important area of research today. In previous work, the CNT-PDMS haptic sensor was used in-sensor based reservoir computing application used in robot hand and achieved an accuracy of >80% for each object when classifying nine different objects[1]. This study focuses on real-time object classification using the CNT-PDMS sensor placed on each finger, as shown in Fig. 1. CNT-PDMS piezoresistive sensors were fabricated by quoting to previous reference[1]. These sensors are simple to manufacture and are produced by adding a small amount of CNT to silicon materials without requiring advanced equipment. The feature of this sensor is that its conductivity increases when pressure is applied. Thus, it will generate different types of signals when grasped different objects. As shown in Fig. 1, the classification process was performed by grasping each of the bottle, cube, and toy strawberry objects 50 times and placing them on four fingers attached with CNT-PDMS sensors (four outputs from each finger), resulting in 16 outputs. The classification process was performed by grasping each of the bottle, cube, and toy strawberry During the remaining intervals, the sensor was in a released state. As shown in Fig. 2a, object grasping led to increased conductivity, resulting in higher voltage values. In Fig. 2b, the classification results represented with colored lines: blue for the bottle, red for the strawberry, and yellow for the cube. Each line peaks during its corresponding object's grasping period, which indicates successful classification in real-time. We plan to increase the number of objects in the future.