Control design for a soft exoskeleton in simulation and identification of the role of different sensor modalities

Kiss, Réka and Köves, Áron Boldizsár and Pelyva, Dávid László and Tasi, Benedek József and Földi, Sándor and Cserey, György and Koller, Miklós (2023) Control design for a soft exoskeleton in simulation and identification of the role of different sensor modalities. In: NSF DARE Conference: Transformative Opportunities for Modeling in Neurorehabilitation, 3-4 Mar 2023, Los Angeles (CA, USA).

Preview	Text nsf_dare_ppkeitk_poster_Koller.pdf - Presentation Download (1MB) | Preview
	Text Kiss_et_al__nsf_dare_conf.pdf - Accepted Version Restricted to Registered users only Download (59kB) | Request a copy

Abstract

People living with athetosis suffer from involuntary movement tremors of irregular amplitude and frequency that can severely impact the realization of their voluntary movements. This condition can be enhanced or maintained only partially, by means of conductive therapy, for example. During our collaboration with a special primary school using conductive pedagogy in the habilitation of affected students, a soft exoskeleton is being developed to assist the smoothening of voluntary arm movements. This system would both partially substitute the on-site active work of conductors during habilitation, and at a later stage passively assist the students to accomplish some specific, potentially workplace-related tasks on their own, with the goal of improving their chances in labour-market integration. On one hand, our work consists of the hardware development of a textile-based, tendon-driven soft exoskeleton system, while on the other, we are examining the related high level control problems, for which three observations are presented in this abstract. Our notices originate either from measurements on actual students of our partner institution, or from imitated movements of healthy subjects. Three different sensor modalities were tried altogether: The first one was a vision-based solution. The RGB and depth streams of an Intel RealSense d435i were processed with Nuitrack (v0.35.13) to get the skeleton. The second measurement system contained inertial measurement units (IMUs, 3 x 9DoF, MPU9250). Recorded sites consisted of the shoulder, the upper and the lower arm. Hard and soft iron calibration was done, logged data was fused with Madgwick's algorithm in order to get absolute 3D orientations. The third measurement type was surface-electromyography (sEMG), directly measuring the activation of three selected muscles (biceps brachii, long and lateral head of triceps brachii). The patch electrodes were wired to 3 channels of a Thalamic Labs Myo Armband, a consumer-friendly, 8-channel sEMG bracelet. Our first observation was made in relation with the low-amplitude tremors: while the vision-based skeletonizer often smoothed out these perturbations in the spatiotemporal trajectory, the IMUs recorded them clearly. This notice has implications for our more accurate movement recordings in the future. The second observation was made in relation with the simultaneous analysis of sEMG and IMU data: although the direct measurement of muscles (sEMG) allows earlier detection of movement initiation, the use of IMU data helps provide a more robust and accurate measurement of displacement. Both the vision-based and sEMG-based data were imported into a biomechanical simulator (OpenSim v4.3). and paired to an existing arm model (MoBL-ARMS, 7DoF, McFarland et al. after Saul et al.). In the case of sEMG data we defined additional external constraints, in order to examine the possible effects of our developed exoskeleton. Although the autonomy of the dynamical system ceased due to the outer effect (muscles' reaction unknown), to a certain degree we could still accept the simulated modified limb movements. This latter neglect enables the short-term optimization of a simulated exoskeleton, serving as our third observation.

Actions (login required)

Edit Item