Slope-Adaptive Gait Planning for the Powered Lower Limb Prosthesis: GRU-Based Coordination Mapping and Its Application

Abstract

This study aims to develop a real-time, slope-adaptive gait planning method for powered knee-ankle prostheses to improve natural gait coordination in transfemoral amputees across variable speeds and slopes. A lightweight gated recurrent unit (GRU) network is proposed, leveraging motion-lagged hip angle, current hip angle, and slope angle as inputs. The system requires only two inertial measurement units (mounted on the socket and prosthetic foot) and a force sensor (at the ankle joint). Network regularization and hyperparameter optimization were applied to enhance computational efficiency, and subject-specific fine-tuning was used for personalized adaptation. The optimized GRU framework achieved high prediction accuracy with a prediction time of <0.007 s, enabling real-time deployment. Experiments with an amputee participant demonstrated seamless adaptation to variable slopes and speeds, improved gait symmetry, and reduced compensatory behaviors in the residual hip and intact ankle. The proposed method enables natural and adaptive gait coordination for transfemoral amputees using a computationally efficient, sensor-minimized approach. This work advances intelligent prostheses toward more natural locomotion in daily environments by improving adaptability to real-world terrain variations.