Effects of stance control hidden Markov model-based gait phase detection on healthy users of an active hip-knee exoskeleton.

In the past years, robotic lower-limb exoskeletons have become a powerful tool to help clinicians improve the rehabilitation process of patients who have suffered from neurological disorders, such as stroke, by applying intensive and repetitive training. However, active subject participation is considered to be an important feature to promote neuroplasticity during gait training. To this end, the present study presents the performance assessment of the AGoRA exoskeleton, a stance-controlled wearable device designed to assist overground walking by unilaterally actuating the knee and hip joints. The exoskeleton's control approach relies on an admittance controller, that varies the system impedance according to the gait phase detected through an adaptive method based on a hidden Markov model. This strategy seeks to comply with the assistance-as-needed rationale, i.e., an assistive device should only intervene when the patient is in need by applying Human-Robot interaction (HRI). As a proof of concept of such a control strategy, a pilot study comparing three experimental conditions (i.e., unassisted, transparent mode, and stance control mode) was carried out to evaluate the exoskeleton's short-term effects on the overground gait pattern of healthy subjects. Gait spatiotemporal parameters and lower-limb kinematics were captured using a 3D-motion analysis system Vicon during the walking trials. By having found only significant differences between the actuated conditions and the unassisted condition in terms of gait velocity ( = 0.048) and knee flexion ( ≤ 0.001), the performance of the AGoRA exoskeleton seems to be comparable to those identified in previous studies found in the literature. This outcome also suggests that future efforts should focus on the improvement of the fastening system in pursuit of kinematic compatibility and enhanced compliance.