Slippage Classification in Prosthetic Hands with a Spiking Neural Network

Poliarticulated myoelectric upper limb prostheses have greatly enhanced dexterity and integration possibilities to patients’ skeletal, muscular and even neural structures. Every manipulation task has, as elemental conditions, touch and slippage identification. As an alternative to the algorithmic approach to slippage detection, we investigate the implementation of spiking neural network aiming the future integration of tactile feedback into hand prosthesis. To this end, we trained a spiking neural network for identifying slippage events during grasping performed by the IH2 Azzurra robotic hand. A comparison with the results obtained from the authors in a previous work is made. In addition to the inherent advantages of the spiking neural network approach, such as biological plausibility and potential ultra-low power hardware implementation, an accuracy of 100% was achieved in the classification of slip and non-slip events.