Graduation Year

2024

Document Type

Dissertation

Degree

Ph.D.

Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Mechanical Engineering

Major Professor

Rajiv Dubey, Ph.D.

Co-Major Professor

Redwan Alqasemi, Ph.D.

Committee Member

Tansel Yucelen, Ph.D.

Committee Member

Sudeep Sarkar, Ph.D.

Committee Member

Kandethody Ramachandran, Ph.D.

Keywords

Motion Capture, OpenSim, Performance Criteria, Redundant Robotic Model, Weighted Least Norm

Abstract

The goal of this research is to develop a human motion-inspired inverse kinematics algorithm framework specifically designed for a Robotics-Based Human Upper Body Model (RHUBM). This framework offers solutions to challenges in various fields. In humanoid robotics, the framework addresses the problem of unnatural robot movement by enabling the development of motion planning algorithms that incorporate human-like movements. For prosthetics, the framework tackles the challenge of amputee difficulty in learning and controlling prosthetics by providing a user-friendly interface that predicts and visualizes upper limb movements, enabling learning and practice. In rehabilitation therapy, the framework tackles the issue of developing effective training protocols by assisting physiotherapists in creating customized programs based on natural human movements, potentially improving stroke recovery.

The central hypothesis of this research suggests that, compared to heavier joints, humans prioritize the movement of lighter joints and integrate various performance criteria—including manipulability, velocity ratio, and mechanical advantage—while avoiding joint limits across different phases of Activities of Daily Living (ADLs). Based on this idea, to develop a framework for motion planning, this research involves five primary objectives, each with specific outcomes.

  • Objective 1 is the creation of a subject-specific, ten-degree-of-freedom RHUBM using a Motion Capture System (MoCap) in the OpenSim and MATLAB software platforms. The MoCap data includes eight range-of-motion (RoM) tasks and nine ADLs, recorded using an eight-camera Vicon motion analysis system. OpenSim enables the calculation of joint angles, which can then be integrated into the robotic model created within MATLAB.
  • Objective 2 is evaluating the influence of arm configurations moving in a null space on each performance criterion and understanding how their combination governs motion.
  • Objective 3 is identifying dominant performance criteria. The full task trajectory is segmented into 4-5 major phases, depending on the specific ADL task. Motion capture data are analyzed to identify the dominant performance criteria during each phase of the respective task. A weight matrix is derived, assigning weights to each joint based on these observations.
  • Objective 4 involves developing an inverse kinematics control algorithm, utilizing the Weighted Least Norm (WLN) approach. This algorithm allocates weights to each joint based on the identified performance criteria, thereby establishing priority among redundant joints and generating human-like motion. The weight matrix used in the WLN algorithm is built on three key components:
  • The first component is Joint Load Management. This component results in less movement of heavier joints, such as the torso, by assigning higher weights to them. This approach minimizes overall joint and muscle strain by reducing movement in heavier joints compared to lighter ones.
  • The second component is Joint Limit Avoidance. This component ensures that joint movements stay within predefined boundaries.
  • The third component is Performance Criteria Optimization. Gradient-based weights optimize specific performance criteria during different task phases, enhancing the overall performance of the robot.
  • Objective 5 involves comparative analysis of WLN algorithm outcomes against MoCap. The MoCap datasets are used as a benchmark to compare the results generated by integrating various weights into the WLN algorithm. The algorithm's efficiency is also xassessed by comparing it against the Least Norm solution. The results indicate that combining link weights, joint limit criteria weights, and gradient-based weights enhances accuracy in recreating human-like motion.

Building upon these results, future research will expand in two stages. The initial stage will focus on expanding the data collection through the recruitment of a more diverse group of participants and the inclusion of a broader spectrum of ADLs. Such expansion is essential for the development of more reliable motion planning algorithms. The subsequent stage aims to investigate various methods for resolving redundancy, including the implementation of advanced machine learning techniques, such as reinforcement learning. These techniques will be tailored to the unique characteristics of users and will prioritize human-like motions. These advancements have the potential not only to enhance the lives of individuals relying on prosthetics and improve the capabilities of robotic assistants but also to facilitate more collaborative and efficient humanrobot partnerships.

Download

Included in

Robotics Commons

Share

COinS