Adaptive SNN for Anthropomorphic Finger Control

Abstract

Anthropomorphic hands that mimic the smoothness of human hand motions should be controlled by artificial units of high biological plausibility. Adaptability is among the characteristics of such control units, which provides the anthropomorphic hand with the ability to learn motions. This paper presents a simple structure of an adaptive spiking neural network implemented in analogue hardware that can be trained using Hebbian learning mechanisms to rotate the metacarpophalangeal joint of a robotic finger towards targeted angle intervals. Being bioinspired, the spiking neural network drives actuators made of shape memory alloy and receives feedback from neuromorphic sensors that convert the joint rotation angle and compression force into the spiking frequency. The adaptive SNN activates independent neural paths that correspond to angle intervals and learns in which of these intervals the rotation the finger rotation is stopped by an external force. Learning occurs when angle-specific neural paths are stimulated concurrently with the supraliminar stimulus that activates all the neurons that inhibit the SNN output stopping the finger. The results showed that after learning, the finger stopped in the angle interval in which the angle-specific neural path was active, without the activation of the supraliminar stimulus. The proposed concept can be used to implement control units for anthropomorphic robots that are able to learn motions unsupervised, based on principles of high biological plausibility.

Bio-Inspired Control System for Fingers Actuated by Multiple SMA Actuators

“…The SNN is based on a neuron model that was previously presented in [ 16 ]. Here we focus only on the critical elements that are used to drive the SMA actuators according to the input signals.…”

“…During each activation of the SOMA, all synapses connected to the hardwired axon generate a spike at their output . The spike energy depends on its amplitude and duration which varies according to the synaptic weights stored by the synapses using capacitors [ 16 ]. In this work all synapses are excitatory and the weights are set to their maximum values to ensure the fastest response of the SMA actuators.…”

“…Note that the input neurons include without INT, the integrator being used only with posts when the switch disconnects . The signal is used for monitoring the neuron activity [ 16 ] because the variation of during neuron activation mimics the postsynaptic membrane potential of the natural neurons.…”

“…Recently, superior abilities of the brain such as symmetry perception [ 7 ], visual pattern recognition [ 8 ], and speech recognition [ 9 , 10 ] were modelled using neuromorphic hardware based on spiking neurons. In robotics, different types of neural systems were designed for the control of vehicles speed [ 11 ] or trajectory [ 12 ], as well as for modelling the motion abilities of the human body including the control of robotic arms [ 13 , 14 , 15 ] or anthropomorphic fingers [ 16 , 17 ].…”

“…In this case, the simulated SNNs use several hundred neurons for the control of each finger of anthropomorphic hands [ 17 ]. In contrast, our previous work demonstrates that less than 50 neurons implemented in analogue hardware are able to control the rotation of a single joint robotic finger actuated by a shape memory alloy (SMA) actuator [ 16 ]. Based on this control principle that was demonstrated for the single actuator, in the current research we evaluate the SNN’s ability to control in sequence the multiple SMA wires that actuate all junctions of the index finger.…”