Bioinspired SNN for robotic joint control

Abstract

In order to implement robotic hands that mimic the smoothness and accuracy of the human hand motions, the artificial control units should be bioinspired. Among the characteristics of such control units is the adaptability which provides the anthropomorphic hand the ability to learn. This paper presents the structure and evaluation of a basic spiking neural network that is able to control the contraction of a SMA actuator using the feedback from a force sensor. Based on this structure we designed an adaptive SNN that can learn to stop the arm in target positions using a sensor that converts the joint rotation angle into spiking frequency. The simulation results showed that the SNN is able to regulate the force of the SMA by balancing the excitatory and inhibitory activity. Also, the force level can be easily set by adjusting the threshold where the inhibitory neurons start to activate.