Chair: Robert Gregg

Abstract:
This thesis introduces a novel control strategy for powered prosthetic legs, designed to enhance their adaptability and functionality across a wide range of daily locomotions such as level walking, ramp walking, stair ascending, stair descending, and the transitions between these activities. Central to this work is addressing three pivotal challenges: creating a tuning-free, biomimetic control system that accommodates multiple activities, adapting to environmental constraints, and developing a self-contained, accurate, and intuitive intent recognition algorithm.

The thesis starts with modeling the able-bodied steady-state kinematics in walking and stair climbing as a continuous function of gait phase, forward speed, and incline. This modeling framework is further refined to incorporate transitions between steady-state activities when led by either the intact or prosthetic leg (i.e., inter-leg transitions) by interpolating between those steady-state kinematics. This approach introduces an innovative method for capturing biomechanical relationships between transitions and the corresponding steady states, which are typically overlooked in prosthetic control. These models have been implemented in a powered knee-ankle prosthesis as reference trajectories to facilitate tuning-free, biomimetic, kinematic control throughout daily locomotion and ensure seamless transitions between them with multiple transfemoral amputee participants.

To improve the adaptability of the powered prosthesis to environmental constraints, the thesis explores the implementation of an advanced stub avoidance controller. This controller utilizes a small, low-cost ultrasonic sensor mounted above the prosthetic ankle to adjust the reference joint kinematics from the aforementioned models in real-time. This method was experimentally validated with amputee participants walking on the powered prosthetic leg, significantly reducing the stub rates when ascending stairs and walking over obstacles of different heights.

Furthermore, the thesis presents an intuitive activity classification system that can automatically switch between the controllers for different activities based on the user’s transition intent using straightforward heuristic rules. This system leverages kinematic features from the thigh and shank angles, along with ground reaction forces and the distance measured by the same ultrasonic sensor, to achieve timely and self-correctable activity recognition. The incorporation of an incline estimation feature further allows the prosthesis to adjust seamlessly between different ground inclines without explicit classifications. This method was experimentally evaluated with amputee participants traversing circuits incorporating daily activities and inter-leg transitions, achieving around 99% accuracy under both self-paced and fast-paced conditions.

The collective findings from this research not only advance the biomechanical functionality of powered prosthetic legs but also facilitate their translation from laboratory settings to practical everyday use. The methodologies and results have been rigorously tested through offline simulations and online experiments involving amputee participants, leading to several publications in top-tier engineering journals and contributions to the field of medical robotics and bionics.

Passcode: 512666