Cerebellum Inspired Neural Architecture for Motor Articulation Control

Abstract

Fast and generalized control executed by brain-inspired models are being employed to design and develop control strategies for robotic arm and neuro-prosthesis. Control strategies from thepast have used the coding capacity of artificial neural networks to predict movements in some robotic platforms. Central nervous system, mainly the brain, has been attributed to stochastic and adaptive behavioural properties for regulating associated functions. Over the years, several newlinehypotheses have been proposed regarding cerebellum and its functioning in motor learning,sparse recoding, coincidence detection and gain adaptation. Sensorimotor association based control can be attributed to the nonlinearity and dynamicity of the underlying microcircuitry yielding timed motor operations and adaptive behaviour. Perceptron like functions that modulate from simple to complex spikes by the activation of granule cell interneuron connections have opened new dimensions in motor learning. Patterns encoded by granular layer neurons also play a major role in adapting Purkinje cell to execute appropriately timed motor behaviours. newlineLearning in neuronal networks has been considered critical for adaptation in complex motortasks. Optimisation of sensory and motor errors that aid in motor movement can be utilised to further engineer human-like tasks such as kinematics and other dynamical problems towards the design of a robust motor controller. With robotics and data science as focus, a cerebellar cortex inspired neural network was reconstructed towards understanding pattern classification, kinematic transformation and other application related to robotics and neuroscience. newlineIn this study, physiologically validated spiking neurons models were used to reconstruct a cerebellum-like architecture to inexplicitly model forward and inverse kinematics and was tested on a low-cost arm. Encoding of input was performed using 3 schemes on a generated robotic dataset which was then provided directly to the spiking neural networks. newline