Multi-Detection Electronic Brush

With the development of the Internet of Things (IoT) society, demand for a variety of sensors is increasing to collect many datasets for data analyses. Among them, whiskers or brush-type sensors have possibilities to contribute to the application of industry and robotics. Brush-type sensors can be used to detect the surface morphology and target shape by contacting the object like animal whiskers, which should be useful for the robotic applications. However, the sensor that can measure such whisker/brush movement has yet to be fully developed. Although some electronic whisker (e-whisker) concepts have been reported previously, a lot of challenges to detect the movement of whiskers/brushes still remain such as multi-detections including movement amplitude, speed, and directions etc. To overcome these challenges, this study proposed out-of-plane e-brush combined with 4 resistive tactile pressure sensors and machine learning system. E-brush is designed to collect multiple datasets precisely from the target. Due to movement of e-brush, tactile pressure sensors show different outputs that allow to distinguish bending amplitude, moving direction, and moving speed simultaneously. Furthermore, reservoir computing (RC), which is one of the recurrent neural networks, was introduced to make real-time data analyses, and it enables to classify 3 brush motion state (stop, move, and slip) and surface condition of contacting object.<br/>First, fundamental characteristics of the sensor were conducted. The brush on the tactile pressure sensor was moved from left to right with applied force from top, repeatedly. During the movement, resistance changes of 4 sensors were simultaneously measured. The tactile pressure sensors can measure a relatively wide range of pressure from 20 Pa to 5.5 kPa. The sensitivity is ~0.33 %/Pa from 0 to 190 Pa and ~0.002 %/Pa from 190 Pa to 5.5 kPa. Due to wide dynamic range and high sensitivity especially at low pressure range, integrated 4 tactile pressure sensors clearly show the output difference by changing the movement velocity (0~90 mm/s), bending distance (-7.5 ~ 7.5 mm), and directions (8 directions). In addition, applying force from top, states of brush movement and surface conditions of contacting object were detected by observing small change of sensor outputs. These results suggest that this platform can measure detail bending information like animal whiskers.<br/>To abstract a lot of information from sensor signals, real-time and quick data analysis system using RC was adapted. By optimizing the RC algorithm, bending distance (-7.5~7.5 mm from the initial position), movement velocity (0~90 mm/s) and force from top (0~1.5 N) were successfully predicted with normalized mean absolute error &lt;0.12 for bending amplitude, &lt;0.20 for velocity, and &lt;0.24 for force. This system also classified movement direction (8 directions), brush motion state (stop, move, and slip), and surface condition (acrylic plate with/without rough sticky tape) with accuracy of ~99 % for direction, ~86 % for motion states, and ~ 98 % for surface conditions. Importantly, this system allows to detect slip conditions of brush structure, continuous movements of brush can be tracked. By integrating all functionalities, as a proof-of-concept, handwriting and its conditions during writing using this brush could be digitized successfully.<br/>In summary, this study demonstrated the e-brush system using integrated tactile pressure sensor and reservoir computing. By optimizing the sensor design and data processing algorithm, the e-brush can detect a variety of datasets, which can be potentially used for the robotic applications.