Identification of slippage on naturalistic surfaces via wavelet transform of tactile signals

The effort toward replicating human skills into artificial systems is growing constantly. While artificial vision has reached a certain reliability, the sense of touch is still hard to introduce into robotic devices. Human manipulation comprises a sequence of static and dynamic actions, which may include unforeseen events, such as variation of object position, movement of the fingers, and modification of the object dimensions and shape (e.g., with soft objects) due to inappropriate force levels. These circumstances are likely to produce the slippage of the object being manipulated. Artificial manipulators are not yet able to be effective in dynamic environments. This paper intends to provide a method for the identification and prevention of slippage with tactile sensors. The method is based on filtering the tactile signals to extract slippage information. The filtering has been executed by means of the stationary wavelet transform that consists of recursive filtering operations. Then, the transformed signal has been rectified and its root mean square has been computed. Finally, an on/off signal has been generated according to a threshold logic. Eight natural surfaces, featuring diverse tactile properties, have been used with the aim of validating the ability of the method to be applied regardless the surface properties. To evaluate repeatability and generalization ability, a total of 2000 experiments have been performed, 250 per each stimulus, with a mechatronic platform: five velocities combined with five indentation force levels, repeating each combination 10 times. Results are provided in terms of true positive detection and of delay between onset of slippage and algorithm output.