A Terrain-based Gait Self-adjusting Planning for Powered Prostheses

Abstract

Lower limb amputees wearing powered prostheses walk not only on flat ground in daily life, but also on complex terrain such as grass and sand. Amputees can reach faster speeds without the need to raise their feet very high on cement roads. While walking on the grass, they are often not as fast, and in order to avoid kicking obstacles, they will raise their feet even higher. We propose a solution that includes terrain recognition based on convolutional neural networks and self-adjusting method of gait parameters to help lower-limb amputees wearing powered prostheses to better adapt to different terrains. Results show, the terrain recognition scheme based on convolutional neural networks has an accuracy of about 95% for the three classification recognition of cement ground, grass, and sand. The trajectory generation method based on self-adjusting gait parameters utilize cubic spline interpolation to generate the desired joint trajectory, thus achieve gait self-adjustment.