Dynamic model refining and identification for accurate gait control of a powered knee–ankle prosthesis

The knowledge of an accurate dynamic model is of fundamental importance for implementing model-based control. However, the model parameters of existing prostheses usually rely on the designed parameters or estimation from single-joint static experiments, which cannot precisely reflect the contribution of inertia, friction, and joint couplings. Taking our newly developed powered knee–ankle prosthesis as an example, we investigated the effect of dynamic identification for model-based control on improving gait accuracy. First, the dynamic model is refined by considering the nonlinear transmission ratio, joint friction, and joint coupling. Then, the static and dynamic identifications were both carried out to illustrate their differences in model validation. At last, a bench test and a walking experiment of gait control with model compensation were made on the prosthesis. The experiments show that the control with dynamic model compensation reduces the position error by 93.44% and the standard deviation of the trajectories of the ankle and knee joints by 31.25% and 44.19%, respectively, compared with the direct control. This result demonstrates the significance of dynamic model identification in the practical application of prosthetic gait control.