Toward a Framework for Adaptive Impedance Control of an Upper-limb Prosthesis

TL;DR

This paper introduces a novel framework enabling upper-limb prosthesis users to simultaneously control kinematics and impedance properties like stiffness and damping, enhancing interaction adaptability and stability during dynamic tasks.

Contribution

The framework integrates muscle-tendon models, forward dynamics, and variable impedance control to allow real-time adaptation of limb properties based on sEMG signals, which is a novel approach in prosthesis control.

Findings

Outperforms baseline in adapting to external perturbations

Improves controllability and user feedback

Demonstrated with both able-bodied subjects and an amputee

Abstract

Adapting upper-limb impedance (i.e., stiffness, damping, inertia) is essential for humans interacting with dynamic environments for executing grasping or manipulation tasks. On the other hand, control methods designed for state-of-the-art upper-limb prostheses infer motor intent from surface electromyography (sEMG) signals in terms of joint kinematics, but they fail to infer and use the underlying impedance properties of the limb. We present a framework that allows a human user to simultaneously control the kinematics, stiffness, and damping of a simulated robot through wrist's flexion-extension. The framework includes muscle-tendon units and a forward dynamics block to estimate the motor intent from sEMG signals, and a variable impedance controller that implements the estimated intent on the robot, allowing the user to adapt the robot's kinematics and dynamics online. We evaluate our framework with 8 able-bodied subjects and an amputee during reaching tasks performed in free space, and in the presence of unexpected external perturbations that require adaptation of the wrist impedance to ensure stable interaction with the environment. We experimentally demonstrate that our approach outperforms a data-driven baseline in terms of its ability to adapt to external perturbations, overall controllability, and feedback from participants.