Real-Time Surface Compliance Detection For Robotic Ankle Prostheses Via Kinematic Data

• Benjamin Chen, Mechanical Engineering, Carnegie Mellon University

• Panagiotis Artemiadis, Mechanical Engineering, University of Delaware

The advancement of robotic ankle prostheses has opened new avenues for improving the mobility and quality of life for individuals with lower-limb amputations. Current powered ankle prosthetic devices are designed to address walking over rigid surfaces, but for people with lower-limb impairments, walking on non-rigid, compliant surfaces can significantly impact their walking stability and increase the risk of falls in this population. Previous studies have explored various methods for surface detection to ensure prosthetic adaptability on a new surface, but the integration of real-time kinematic data from the prosthesis to predict the compliance of the upcoming surface has not yet been addressed. Our research aims to enhance the adaptability and responsiveness of robotic ankle prostheses during transitions from rigid to compliant surfaces by implementing a real-time classification framework to predict changes in the surface stiffness. This work proposes a Support Vector Machine (SVM) classification model that utilizes only the kinematic data provided by the ankle prosthesis to predict the upcoming surface stiffness in real time and send a signal to the prosthesis, enabling the device to dynamically adjust the prosthesis parameters to suit the detected surface. After data analysis, we determined that the ankle angle, ankle moment, ankle velocity, and the rotational motion of the ankle prosthesis in the x-axis provided the optimal input for high classification accuracy. Upon training our classification framework on data collected from users walking with the prosthesis on a unique robotic device, the Variable Stiffness Treadmill (VST) which allows us to simulate walking on both rigid and compliant surfaces, we achieved a 90% accuracy within tenths of a second after each heel strike. The proposed classification model can therefore reliably predict the upcoming surface stiffness with high accuracy, demonstrating that real-time surface classification can greatly enhance the functionality and stability of robotic ankle prostheses.