Pattern classification of terrain during amputee walking

Abstract

In this thesis I study the role of extrinsic (sensors placed on the body) versus intrinsic sensing (instruments placed on an artificial limb) and determine a robust set of sensors from physical and reliability constraints for a terrain adaptation in a robotic ankle prosthesis. Further, during this thesis I collect a novel data-set that contains seven able-bodied participants walking over 19 terrain transitions and 7 non-amputees walking over 9 transitions, forming the largest collection of transitions to date using an exhaustive set of sensors: inertial measurement units, gyroscopes, kinematics from motion capture, and electromyography from 16 sites on the lower limbs for non-amputee subjects and 9 sites or amputee subjects. This work extends previous work [3] by using more conditions, a larger subject group, and more sensors on amputees, and includes non-amputees.I present a novel machine learning algorithm that uses sensor data during rapid transitions from pre-foothold to just prior to post-foothold to predict different terrain boundaries. This advances the field of biomechatronics, our understanding of terrain adaptation in people both with and without amputations, contributes to the development of a fully terrain adaptive robotic ankle prosthesis, and improves the quality of life for the physically challenged. Specifically we set out to prove between pre and post-foothold the ankle and knee positions calculated using an IMU attached to an amputees powered prosthetic ankle can discriminate with greater than 99% accuracy between 9 conditions. Our results suggest that myography as a non-volitional sensing modality for terrain adaptive prostheses was not needed.