Machine Learning for Control
of Soft Exoglove

depositphotos_90001218-stock-illustratio

M. Sierotowicz, N. Lotti, L. Nell, F. Missiroli, R. Alicea, M. Xiloyannis, R. Rupp, E. Papp, J. Krzywinski, C. Castellini and L. Masia,

EMG-driven Machine Learning Control of a Soft Glove for Grasping Assistance and Rehabilitation”.

IEEE Robotics and Automation Letters (RA-L) , vol. 7, no. 2, pp. 1566-1573, April 2022, doi: 10.1109/LRA.2021.3140055. PDF

In the field of rehabilitation robotics, transparent, precise and intuitive control of hand exoskeletons still represents a substantial challenge. In particular, the use of compliant systems often leads to a trade-off between lightness and material flexibility, and control precision.

This project aims at proposing a compliant, actuated glove with a control scheme to detect the user's motion intent, which is estimated by a machine learning algorithm based on muscle activity.

glovewebsite.png

 

Figure 1 MyoGlove design and control. a: the presented prototype actively assists hand grasping by means of tendon-driven actuators (1) lodged in the driver's stage. Hand opening is passively aided by three 3D printed elastic straps integrated in the glove (2). Hand movements are sensed by means of two flex sensors connected to a Bluetooth data collection module measuring overall thumb and index flexion/extension (3-4). The motion intent is estimated by means of sEMG  (5). A magnetic clutch has been developed in order to allow for quick (de)coupling of the glove from the driver stage allowing for quick adjustments to the user's hand size (visible in the lower right corner of a).  b: block diagram depicting the control architecture of the MyoGlove. The high-level controller aims to estimate the motion intent through a ridge regression or a random Fourier features algorithm. This estimate is then used as set-point and compared with the angular velocity measurement of the flex sensors in the low-level admittance controller which translates the tracking error into a motor angular velocity, which is finally converted into an actuation command.

 

Participants used the glove in three assistance conditions during a force reaching task. The results suggest that active assistance from the glove can aid the user, reducing the muscular activity needed to attain a medium-high grasp force, and that closed-loop control of a compliant assistive glove can successfully be implemented by means of a machine learning algorithm.