Robust and Adaptive Lower Limb Prosthesis Stance Control via Extended Kalman Filter-Based Gait Phase Estimation

We present a control strategy for powered prostheses based on a robust estimate of the gait phase that is used to determine appropriate control actions. We use an extended Kalman filter (EKF) that fuses joint angle and velocity measurements to estimate the gait phase, which we define in this work to be a variable that progresses continuously during stance from zero at heel strike to one at toe-off. The control strategy uses the gait phase estimate as the input into Gaussian process (GP) functions that specify the desired angles, velocities, and feed-forward torques for the knee and ankle joints. We compare this proposed GP-EKF control strategy to two alternative controllers: A neuromuscular (NM) control strategy, which models leg muscles and hypothesized reflexes, and an impedance (IMP) control strategy. Our experiments involved seven able-bodied participants and a single amputee participant. We find that the GP-EKF control generated knee angle trajectories that were significantly closer to able-bodied walking data than those produced by either the NM or IMP controllers. However, ankle trajectories were less similar. In addition, we find in experiments with the participants stepping on blocks during stance that the GP-EKF control resulted in significantly fewer fall-like events than IMP control. Finally, we evaluate the ability of the proposed control to track the gait phase across both slowly and rapidly varying treadmill speeds and find that the EKF's phase estimate tracked these gait changes significantly better than a time-based phase estimate. The proposed control strategy may provide a robust and adaptive control alternative for powered prostheses.