Decoding invasive and non-invasive human brain signals for robotic arm control using machine learning techniques through brain-machine interface (BMI)

Abstract

Brain-computer interface is a useful technology that describes a framework to decode human neural signals to understand user intentions that may be translated to actions commands for controlling external output devices. To this end, previous work in the past decades have conducted experiments on human and non-human primate subjects to explore the possibility of using neural signals to control assistive devices that may aid the user in daily living, and these attempts were often met with high success. Among these breeds, the brain-computer interface for robotic arm control has gained much attention in recent years. A robotic arm is able to assist users to perform tasks that may complement the normal functioning of the healthy arm, or can be used to assist patients with motor impairments such as tetraplegia by replicating the functionality of the healthy arm that may be needed to perform simple and complex tasks under real-life settings. In this thesis, the decoding of invasive and noninvasive human neural signals during hand movements to reach-and-grasp objects was investigated to determine whether trajectory of intended hand movements could be decoded accurately, and explored the possibility of developing this paradigm to design a brain-computer interface for robotic arm trajectory control.