Abstract
The presence of a tactile sensor is essential to hold an object and manipulate it without damage. The tactile information helps determine whether an object is stably held. If a tactile sensor is installed at wherever the robot and the object touch, the robot could interact with more objects. In this paper, a skin type slip sensor that can be attached to the surface of a robot with various curvatures is presented. A simple mechanical sensor structure enables the cut and fit of the sensor according to the curvature. The sensor uses a non-array structure and can operate even if a part of the sensor is cut off. The slip was distinguished using a simple vibration signal received from the sensor. The signal is transformed into the time-frequency domain, and the slippage was determined using an artificial neural network. The accuracy of slip detection was compared using four artificial neural network models. In addition, the strengths and weaknesses of each neural network model were analyzed according to the data used for training. As a result, the developed sensor detected slip with an average of 95.73% accuracy at various curvatures and contact points.
Highlights
The sense of touch is a handy tool when picking up or manipulating objects
This paper presents a sensor that detects slips effectively by combining an artificial neural networks (ANN) and a simple sensor structure
A metalized piezo film sheet (2-1004347-0, Measurement Specialties, Hampton, VA, USA) was used. This film has a 52 μm-thick polyvinylidene fluoride (PVDF) layer polarized in the direction of thickness and has electrodes formed on both sides of each 6 μm-thick silver ink
Summary
The sense of touch is a handy tool when picking up or manipulating objects. Checking the hardness of an object, distinguishing the texture of the surface, and determining the degree of friction are very difficult without the sense of touch. Sensors that acquire higher-level information using signal processing are being developed [5] They measure the normal force, shear force, and vibration of the surface to distinguish the material or detect slip on objects. Li et al processed the response and vision information of GelSight sensor with DNN to determine the slip with an accuracy of 88.03% in 10 unseen objects [13]. These ANNs make it possible to use simple and inexpensive sensors. After extracting features from the measured sensor data, the slip was detected using four ANN of different structures.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
