Autonomous dexterous manipulation relies on the ability to recognize an object and detect its slippage. Dynamic tactile signals are important for object recognition and slip detection. An object can be identified based on the acquired signals generated at contact points during tactile interaction. The use of vibrotactile sensors can increase the accuracy of texture recognition and preempt the slippage of a grasped object. In this work, we present a Deep Learning (DL) based method for the simultaneous texture recognition and slip detection. The method detects non-slip and slip events, the velocity, and discriminate textures—all within 17 ms. We evaluate the method for three objects grasped using an industrial gripper with accelerometers installed on its fingertips. A comparative analysis of convolutional neural networks (CNNs), feed-forward neural networks, and long short-term memory networks confirmed that deep CNNs have a higher generalization accuracy. We also evaluated the performance of the highest accuracy method for different signal bandwidths, which showed that a bandwidth of 125 Hz is enough to classify textures with 80% accuracy.

We describe an algorithm that can robustly decide whether a grip or a footstep is secure given data collected from at least two independent sensors. This algorithm is based on the observation that if there is an absence of slip, then, owing to the high velocity of mechanical waves in solids, the two sensor signals must be highly correlated, even in the presence of internal or external perturbations. The statistical distance between signals collected during slip and non-slip phases, regarded as random distributions, also provides a continuous measure of graspability or walkability of an object being held or a ground being stepped on. We tested the algorithm on a bench using micro-electro-mechanical system (MEMS) accelerometers and with a variety of materials of different surface roughnesses. We also discuss the applications of this non-slip/slip discrimination algorithm and its putative relationship with human gripping behavior.

Through vibrations people can assess both slippage of the directly grasped object and sliding of an external agent over the surface of the grasped immovable object. In robotic hands, with less advanced tactile sensing, the slippage and sliding events can be hard to distinguish. This paper shows how vibro-tactile sensing array can help to distinguish object/world sliding and sensor/object slippage events based on cross-correlation, which computes similarity in sensor readings of tactile array cells. The proposed vibro-tactile system consists of two accelerometers. Experiments with different surfaces are conducted to test the system and the proposed algorithm.

Granular particles, filled within an elastic material, produce mechanical vibrations in structures or air when squeezed. This refers to structure-borne noise, is defined as a noise that occurs from the impacts of particles hitting each other due to their momentum. The momentum depends on both properties of particles and velocity of squeezing. Therefore, the structure-borne noise is highly correlated with the properties of particles. In this connection, we study a vibro-tactile sensor for detecting the mechanical vibrations from squeezing granular objects. Specifically, we explore machine learning solutions to detect foreign body within these objects using detected vibrations. We evaluated multiple learning approaches on a collected data set of 900 squeezing experiments across 15 different granular materials. In our experiments, the most successful method was convolutional neural network that achieved an accuracy of 91% on unseen test data. Remarkably, the foreign body was detected with a higher success rate for the majority of material types except salt and coffee granules.

No abstract