Smart Step: Wearable Mobility Assistance using Machine Learning and Haptic Feedback

Abstract

Reinventing the sensory world for prosthesis users remains a challenge. I leverage recent advances in machine learning and miniaturized sensors to build a smart wearable that tackles a cognitively demanding and ill-researched mobility task: stair descent. The human sensory system enables error correction of foot placement during stair descent; however, amputation removes access to much of this information. People with lower limb amputations often compensate for the absence of ankle range of motion and plantar sensation by employing the “overhanging toe” strategy during stair descent. Such users will pivot the foot over the stair edge in a smooth rolling motion, which requires the prosthetic foot to be placed optimally on the stair edge. Smart Step replaces the role of the human sensory system in foot placement, particularly stair descent. Smart Step consists of the sensing element (insole with force sensors and wearable Inertial Measurement Units strapped on the body) and the cueing element (haptic band worn on the thigh or the wrist). The sensors will feed data into a machine learning algorithm, predicting foot placement, and subsequently provide instructional haptic cues for the user. The cues enable users to achieve optimal foot placement with each step, thus the device adopts the cognitive load and enables a smooth and controlled gait during stair descent. By using low cost materials and noninvasive techniques to fit in any shoe, Smart Step may be useful not only for prosthesis users, but also for anyone with gait impairments.