Detection of incipient object slippage by skin-like sensing and neural network processing

Abstract

Detection of incipient slippage is of great importance in robotics for the control of grasping and manipulation tasks. Together with fine-form reconstruction and primitive recognition, it has to be the main feature of an artificial tactile system. The system presented here is based on a neural network used to detect incipient slippage and on a skin-like sensor sensible to normal and shear stresses. Normal and shear stresses components inside the sensor are the input data of the neural net. An important feature of the system is that the a priori knowledge of the friction coefficient between the sensor and the object being manipulated is not needed. To validate the method we worked on both simulated and experimental data. In the first case, the finite element method is used to solve the direct problem of elastic contact in its full nonlinearity by resorting to the lowest number of approximations regarding the real problem. Simulation has shown that the network learns and is robust to noise. Then an experimental test was carried out. Experimental results show that, in a simple case, the method is able to detect the insipiency of slippage between an object and the sensor.

Tactile Sensors for Friction Estimation and Incipient Slip Detection—Toward Dexterous Robotic Manipulation: A Review

“…7), and a NN was trained to detect incipient slip. Canepa et al [74] described a tactile sensor consisting of a linear array of eight pairs of piezoelectric polymer transducers underneath a layer of silicone rubber. One transducer in each pair is sensitive to normal stress and the other to shear stress.…”

“…Without that validation, there is a question whether the sensor is in fact detecting incipient slip, or whether it is simply able to detect the gross slip earlier than other traditional reference methods, which may be the case in [72,85,86]. Only one of the works has attempted to validate the incipient slip detection itself [74], and a few works have validated the incipient slip detection against simulated results [75][76][77]88]. That being said, a number of these works [75-78, 81, 83, 86, 91, 92] have attempted to integrate their sensors into some kind of gripping feedback control, which give a form of indirect validation.…”

Neural Network Training Techniques for Enhancing the Performance of a Distributive Tactile Sensor

“…When integrated with a robot finger or multiple fingers, tactile sensing can be devised to measure grasped object profiles [1][2][3], identify object materials [4] and determine contacting load and location [5]. Some applications have been devised for controlling and manipulating grasped objects; for example, devices for detecting slip [6][7][8]. Brett and Stone [9] and Evans and Brett [10] examined an approach to determine a contact with soft objects and for adjusting a grasping strategy to control deformation.…”

Grasping force control in consideration of translational and rotational slippage of an object by a flexible contact sensor

“…Ikeda and colleagues [6] employed the relations between shear deformation and slip/stick regions within the contact area based on the Hertz contact theory [5] to derive equations that do not contain the friction coefficient explicitly; as a result, they achieved slip-free grasping with unknown friction coefficients. Canepa and colleagues [7] used the FEM to analyze the stress distribution pattern of incipient slip between the sensor and object; they proposed slippage prevention based on such distribution patterns, using a combination of low-resolution tactile sensors and a neural network. In addition, Shinoda and colleagues estimated the friction coefficient from the strain variation in the contact area during contact with a flexible sensor [8].…”