Feature Matching-Based Gait Phase Prediction for Obstacle Crossing Control of Powered Transfemoral Prosthesis

TL;DR

This work addresses obstacle crossing for powered transfemoral prostheses by predicting prosthetic hip and knee trajectories from sound-leg ankle height data using a neural network whose architecture is optimized by a genetic algorithm. A DTW-based feature-matching approach then estimates gait progression within the obstacle-crossing phase to determine the knee actuation index. The method achieves knee and thigh RMSEs of $6.78\%$ and $8.71\%$, with gait-phase accuracy reaching $100\%$ up to $150$ Hz and robustness to Gaussian noise with SD below $1$. These results suggest a practical, data-driven pathway for reliable obstacle negotiation in real-time prosthetic control, pending outdoor validation with amputee subjects.

Abstract

For amputees with powered transfemoral prosthetics, navigating obstacles or complex terrain remains challenging. This study addresses this issue by using an inertial sensor on the sound ankle to guide obstacle-crossing movements. A genetic algorithm computes the optimal neural network structure to predict the required angles of the thigh and knee joints. A gait progression prediction algorithm determines the actuation angle index for the prosthetic knee motor, ultimately defining the necessary thigh and knee angles and gait progression. Results show that when the standard deviation of Gaussian noise added to the thigh angle data is less than 1, the method can effectively eliminate noise interference, achieving 100\% accuracy in gait phase estimation under 150 Hz, with thigh angle prediction error being 8.71\% and knee angle prediction error being 6.78\%. These findings demonstrate the method's ability to accurately predict gait progression and joint angles, offering significant practical value for obstacle negotiation in powered transfemoral prosthetics.

Figures (6)

• Figure : Fig. 1. A simplified explanation of the implementation steps of the method. Section II-A explains the collection and processing of experimental data, Section II-B describes how to train and find the best neural network for predicting hip and knee joint angles also with gait phase estimation, Section II-C explains how to predict gait phase (gait progression), and finally Section II-D briefly describes how to convert the calculated angles into motor torques.
• Figure : Fig. 2. Data collection settings. (a) An IMU is attached to the lateral side of the subject's right ankle. Motion capture markers are positioned at three key points: the most anterior point of the iliac crest, the most lateral aspect of the femoral epicondyle, and the outermost point of the ankle joint. (b) This Depicts the specific gait strategy planned for the subject, with red tape indicating the landing points for the subject's feet, and the obstacle composed of sponge.
• Figure : Fig. 3. Hip and knee angle correspondence during obstacle crossing. The magenta-colored angle data represents the data from the phase prior to the obstacle-crossing stage, while the green-colored angle data corresponds to the phase during the obstacle-crossing stage, which is the target data for prediction.
• Figure : Fig. 4. Optimal Network Search Block Diagram. This illustrates obtaining the optimal network structure via a genetic function to optimize both gait cycle progress and joint angle trajectory predictions. "Angle height" denotes the height data of the healthy leg's ankle during obstacle crossing, while "Pre-thigh angle" refers to the angle data of the prosthetic leg before crossing. "Network structure" is the architecture passed from the genetic function to the network training function. "Thigh angle" represents the predicted thigh angle generated by the network. "Error1," "Error2," and "Error" correspond to the prediction errors of gait cycle progress, hip and knee joint angles, and their weighted sum, respectively.
• Figure : Fig. 5. Correspondence diagram of predicted angle and actual angle. The two colors from light to dark represent the correspondence of each group of real data and predicted data.