Date of Award

2025

Type

Thesis

Major

Master of Science

Degree Type

Master of Science in Robotics Engineering

Department

Earth & Space Science

First Advisor

Dr. Mohammad Jafari

Second Advisor

Dr. Abiye Seifu

Third Advisor

Dr. Rania Hodhod

Abstract

This thesis develops and experimentally validates an online adaptive Linear Quadratic Regulator (LQR) control method for prosthetic joint systems using Recursive Least Squares (RLS)-based real-time parameter estimation on the Quanser QUBE-Servo 2 platform. Traditional LQR controllers assume a fxed system model, which limits adaptability and results in reduced tracking accuracy, poor robustness, and loss of optimal performance when applied to dynamically changing prosthetic joints, infuenced by load variations, user gait changes, and mechanical wear. Limited experimental validation exists for combining RLS with online LQR adaptation in prosthetic-like systems.

To address these limitations, an RLS-driven Adaptive LQR framework was implemented to continuously update LQR gains in real-time based on measured system parameters. Experiments on two QUBE-Servo 2 systems demonstrated that, compared to the fxed-gain LQR, the Adaptive LQR improved performance for Qube 1 and Qube 2 as follows: mean squared error (MSE) by 36.2% and 69.4%, root mean squared error (RMSE) by 20.1% and 44.7%, mean absolute error (MAE) by 25.0% and 59.8%, and mean absolute percentage error (MAPE) by 25.0% and 59.8%, respectively. The Adaptive LQR also exhibited consistent performance across both platforms, whereas the fxed-gain LQR showed substantial variability between Qube 1 and Qube 2. These results confrm that RLS-driven online adaptation can signifcantly enhance tracking accuracy, robustness, and adaptability, supporting its potential application in prosthetic joint control and other dynamically changing systems.

Download

Share

COinS