Continuous Gait Phase Estimation and Torque Profile Generation Using Adaptive Dynamic Movement Primitives for Able-Bodied Individuals and Stroke Survivors

Continuous gait phase estimation facilitates the real-time synchronization of wearable robotic assistive strategies with the user's movements, ensuring seamless operation across varying walking speeds. A previous study showed that adaptive Dynamic Movement Primitives (aDMP) can estimate the gait phase in real time for various locomotion activities of daily life. Advancing this paradigm, this study has two objectives. Firstly, it investigates the use of aDMP-based gait phase estimation to generate phase-dependent torque profiles using a powered unilateral hip exoskeleton. Tests were conducted with six ablebodied participants in level-ground walking. In these tests, the performance of the aDMP algorithm was assessed using subjectdependent and subject-independent models. Secondly, this study investigates the applicability of aDMP to estimate the gait phase for a pathological gait, using a dataset from six stroke survivors. Results with able-bodied participants demonstrated that root mean square errors (RMSE) in phase estimation were between 2.75-5% relative to the linear phase across all experimental conditions, including assistive and transparent (i.e., zero-torque) modes with subject-dependent and subject-independent models. Results with stroke survivors exhibited a median RMSE of 6.75% when the exoskeleton was operated in transparent mode. These results demonstrated that aDMP-based gait phase estimation can effectively generate phase-locked torque profiles for able-bodied participants. However, further research is needed to validate these findings in stroke survivors, as the preliminary offline results indicated a higher RMSE in this population.