Reference Trajectory Generation for Closed-Loop Control of Electrical Stimulation for Rehabilitation of Upper Limb

• Published in: IFAC-PapersOnLine Vol. 53; no. 2; pp. 16438 - 16444
• Main Authors: Karak, Tarun, Tiwari, Laxmi Kant, Sengupta, Somnath, Nag, Sudip
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 2020
• Subjects: BCI; classification of grasping; closed-loop control; EEG; EMG; FES; machine learning; trajectory generation; FES; classification of grasping; EEG; BCI; closed-loop control; trajectory generation; machine learning; EMG
• ISSN: 2405-8963; 2405-8963
• DOI: 10.1016/j.ifacol.2020.12.710

Functional movements in the paralyzed upper limb can be restored with the help of brain-computer-interface (BCI). A BCI system typically adopts a functional electrical stimulation (FES) system that activates weakened muscles that are otherwise responsible for actuating finger movements. A BCI-FES system can enable muscle contraction through the delivery of electrical stimulation pulses. The control of voltage or current stimulation parameters such as pulse width, frequency, and amplitude along with feedback signals from finger joints positions are essential for stable grasping. For the design of a closed-loop functional electrical stimulation controller, it is obligatory to set standard reference trajectories of finger joints’ angular positions and velocities for controlling stimulation parameters in neuroprosthetics and rehabilitation. This study proposes a new closed-loop control architecture targeted for achieving successful and stable grasping of an upper limb paralyzed subject. This can be achieved by characterizing each of the finger joints’ instantaneous angular position and velocity, through reference trajectories. These reference trajectories are generated corresponding to various types of grasping for feeding to the controller, responsible for stimulation of muscles. Hence, to generate such trajectories, first, grasping classification has been implemented using standard machine learning algorithms on a large set of existing real-time data of different types of objects’ grasping such as various diameter, abducted thumb and other types of objects, from many healthy subjects. The results demonstrate the successful implementation of fairly accurate classifications and trajectory generations which are crucial for closed-loop control towards stable grasping.