Teleimpedance Based Assistive Control for a Compliant Knee Exoskeleton

Abstract

This paper presents a tele-impedance based assistive control scheme for a knee exoskeleton device. The proposed controller captures the user's intent to generate task-related assistive torques by means of the exoskeleton in different phases of the subject's normal activity. To do so, a detailed musculoskeletal model of the human knee is developed and experimentally calibrated to best match the users kinematic and dynamic behavior. Three dominant antagonistic muscle pairs are used in our model, in which electromyography (EMG) signals are acquired, processed and used for the estimation of the knee joint torque, trajectory and the stiffness trend, in real time. The estimated stiffness trend is then scaled and mapped to a task-related stiffness interval to agree with the desired degree of assistance. The desired stiffness and equilibrium trajectories are then tracked by the exoskeletons impedance controller. As a consequence, while minimum muscular activity corresponds to low stiffness, i.e. highly transparent motion, higher co-contractions result in a stiffer joint and a greater level of assistance. To evaluate the robustness of the proposed technique, a study of the dynamics of the human-exoskeleton system is conducted, while the stability in the steady state and transient condition is investigated. In addition, experimental results of standing-up and sitting-down tasks are demonstrated to further investigate the capabilities of the controller. The results indicate that the compliant knee exoskeleton, incorporating the proposed tele-impedance controller, can effectively generate assistive actions that are volitionally and intuitively controlled by the user's muscle activity.

Overground Walking With a Transparent Exoskeleton Shows Changes in Spatiotemporal Gait Parameters

“…Different compliant control strategies were proposed in gait rehabilitation following the assistance-as-needed (AAN) concept [11] . Active impedance control and its variants [12] , [13] , [14] and the teleimpedance approach [15] are some examples. Despite the achieved outcomes, these techniques face limitations in generating suitable net user-robot interaction forces due to the weight and inertia of the system and poor backdrivability of the actuators.…”

Adaptive motion intent understanding–based control of human–exoskeleton system

“…It is widely acknowledged that a musculoskeletal model applicable to human motion prediction is the precondition of realizing transparent human–exoskeleton interaction, 6 since the exoskeleton should have knowledge of the human’s planed action in order to perform efficient assistances. 7,8 Strategies of modeling the musculoskeletal system can be classified into two categories, that is, model-based algorithms and non-model-based methods. 9 Han et al 10 proposed a state space model for the estimation of joint movements.…”

“…Commonly used machine learning methods include neural networks (NNs) 14 and Gaussian process (GP). 3 In Ajoudani, 8 the NN was used to estimate the joint torques. The GP model was established to predict the joint angle when surface electromyography (sEMG) signal was treated as one of the inputs.…”