BCI based Robotic Arm Control using MI-EEG and Spiking Neural Network

Abstract

Brain Computer Interface is a technology that helps to interact with human nervous system to analyze and stimulate neural circuits. This research work focuses on the Motor Imagery - Electroencephalography signal for robotic hand movements capable of lifting and dropping artificial limbs. The signals are acquired from subjects of age group between 35 to 60 upon lifting and dropping their hand. The raw signals were pre-processed with signal processing techniques for noise and artifact removal. Further, 2D-spectrogram features were extracted for training the 2D-Convolutional Neural Network (2D-CNN) and a Spiking Neural network (SNN). SNN algorithm is a third-generation neural network closely resembling the complex neuron structure of the brain, being energy-efficient with lesser iterations and transmits information as discrete spike trains. The best average accuracy achieved through 2D-CNN is 99.06% with raw signal images and through SNN is 97.38% when trained upon spectrogram images. The trained model is later interfaced with Arduino to move the robotic arm.