Identification of Contact and Non-Contact Finger Motion using Surface Electromyography(sEMG): An Explainable AI Approach

Motion-copying systems (MCS) enable the transfer of motor skills from humans to humans or humans to robots. MCS are becoming increasingly important in the context of human-robot interaction (HRI). However, the biggest challenge with MCS is the lack of a standard for motion abstraction. In practice motion abstraction mostly depends on the end use case or the control scheme employed. The recent advancements in control systems such as hybrid and adaptive impedance control demand the simultaneous abstraction of both the force and position information from motion. In addition, it is also critical to classify the motion type i.e. free or forced motion, and identify the instance of contact. Most methods focus on either the position or the force information. Methods that consider both force and position are not scalable, portable, and/or employ a black box model. In addition, these methods of motion abstraction are unable to classify the motion type. Considering all of these challenges this research proposes the use of wearable sensors, surface electromyography (sEMG), and flex sensors for the abstraction of motion directly from muscles and focuses on a novel framework for identifying contact state and instance of contact without the use of any additional sensors. In addition, the proposed framework will enable context-aware control strategies and is a great resource for human-in-the-loop control strategies for applications such as motor rehabilitation and assistive devices.