Towards implementing impedance control in sensorimotor neuroprostheses

Abstract

Neural prostheses, or brain-machine interfaces (BMI) are a series of devices that can substitute a motor, sensory or cognitive modality that might have been damaged as a result of an injury or a disease. Current sensorimotor BMI research uses neural predictions of kinematic variables like position or velocity to drive robotic arm, wheelchair or computer cursor. But use of neurally predicted dynamic variables, like force or torque, to drive a BMI has not been demonstrated. Involving dynamic variables for BMI control is important because simplest reach-to-grasp movements extensively use forces/torques and impedance control, apart from tactile/proprioceptive/visual feedback, for dexterous maneuvering of/around the objects. We used non-human primates (bonnet macaque, M. radiata) to demonstrate BMI movements involving such dynamic signals, because they serve as a very close animal model for human reach-and-grasp movements. Towards achieving the goal of impedance control in non-human primate setup, first we need a method to acquire motor cortical signals with high spatiotemporal resolution in a bio-friendly, reversible, economical and repeatable manner while being frugal on the number of animals used. Secondly, we must find a candidate behavioral signal that is both efficient and generalizable to a variety of external dynamic environments. Finally we have to demonstrate the use of such signal in real-time, closed-loop fashion towards more realistic movements involving forces/torques and ultimately impedance control. This thesis aimed at (1) Developing a microelectrode implantation technique that is bio-friendly, reversible, repeatable and economical; (2) Finding a behavioral signal that can control the BMI consistently across different external dynamic environments; and (3) Developing a real-time control of movements that has improved interaction with the external environment and is capable of impedance control. Towards aim 1, we have developed a method of microelectrode implantation using 'Nesting Platform' and used it successfully on multiple animals. It also worked well with repeated microelectrode implantation on a single animal in chronic setup, making efficient use of an animal that has been trained on the reaching task for months to years. Towards aim 2, we have demonstrated that power, a combination of both kinematic and dynamic variables, gives consistent performance across variety of dynamic environments in offline setup. Towards aim 3, we have demonstrated the use of hybrid torque+position control in real-time, closed-loop manner and that changing relative contribution of each variable leads to change in the admittance/impedance of the virtual arm. Finally, current state, future implications and relevant ethical implications of the neuroprosthetics in general have also been discussed.