Ferrante, Laura ORCID: 0000-0002-9408-3024 (2023). Towards adaptive impedance control for upper-limb prostheses. University of Birmingham. Ph.D.
Ferrante2023PhD.pdf
Text - Accepted Version Restricted to Repository staff only until 3 January 2033. Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (24MB) | Request a copy |
Abstract
This work presents a novel framework that estimates the human motor intent from surface electromyographic (sEMG) signals and executes the motor intent on a simulated robot, allowing the user to adapt the kinematics and dynamics of a Degree of Freedom in static and dynamic environments.
Adaptation of intrinsic mechanical properties of the arm and hand, during the execution of grasping and manipulation tasks, is crucial for humans to physically interact with unpredictable and dynamic environments. These properties are defined as stiffness, damping and inertia, and determine a complex neuromuscular behaviour known as impedance. Consider holding an umbrella and grasping the handhold more firmly to counter the action of the wind blowing. To do this we co-contract flexor and extensor muscles spanning the elbow and the wrist to make the arm more rigid. The action of these muscle groups opposes one another to maintain the arm’s posture, but it also contributes to increasing limb impedance. As a result, humans can regulate the limb’s kinematics and dynamics independently, according to the task requirements.
Inspired by human motor control, researchers have designed biomimetic controllers to enable such impedance behaviour on robotic systems that physically interact with the environment. The most important and challenging requirement of such controllers is to adapt the robot’s impedance over time, according to task and environment constraints. This requirement is desirable in different physical human-robot interaction applications, and crucial in motor prostheses control where the human motor intent has to be decoded and implemented in real-time on the artificial limb. The latter is the application domain that motivates the work presented in this thesis.
Today, none of the commercially available prostheses allow the user to simultaneously control the kinematics and impedance of a single degree of freedom of the robotic arm. Motor prostheses use low-density surface electromyography as a human-machine interface to allow the user to communicate their motor intent through muscle contraction. However, the estimation of the impedance properties from sEMG signals is not trivial, due to the low bandwidth of sEMG signals and due to the complexity and redundancy of the human musculoskeletal system.
Methods that attempt to fill this gap typically use sEMG-driven muscle-tendon models that potentially allow to predict the torque applied at the joint and the evolution of the muscle-tendon models state, from which the stiffness and damping of the joint can be obtained. However, none of the existing methods coherently estimates the joint impedance from these models and uses it to implement a variable impedance controller on the robot. While the muscle-tendon models are used to obtain the intended joint kinematics, the joint stiffness and damping are estimated from other models, usually polynomial functions of the sEMG signals. Moreover, the obtained joint stiffness and damping are not directly used as gains in the control law, but these are remapped to suitable ranges to ensure control stability and satisfy hardware requirements. As a result, multiple calibration phases are required, which prevent a coherent implementation of the user’s motor intent on the robot and may impact the user’s control performance.
This thesis presents an sEMG-driven framework that provides the user with 3 Degrees of Control for a single Degree of Freedom of a simulated robot, actuated through wrist flexion-extension. The framework includes a pair of muscle-tendon models to estimate the motor intent from two sEMG signals in terms of reference joint motion, stiffness and damping. Unlike previous work, the parameter values of the muscle-tendon models are estimated such that the obtained joint stiffness and damping can be used directly to implement a variable impedance controller on the robot. This ensures that the user’s intended dynamics, represented by the muscle-tendon models, matches that of the robot, enhancing the framework’s transparency in implementing the user’s motor intent. As a result, the human subject is able to simultaneously adapt the robot’s kinematics and dynamics on-the-fly.
We evaluate our framework with eight able-bodied subjects during reaching tasks performed in free space, and in the presence of unexpected external perturbations that require adaptation of the wrist impedance to ensure stable interaction with the environment. A case study is carried out with a transradial amputee. The proposed framework is compared to a baseline consisting of a purely data-driven method that learns a mapping from sEMG signals to desired joint kinematics and a fixed-gains high-stiffness controller that tracks the estimated kinematics. We investigate whether our framework, which enables kinematics as well as stiffness and damping adaptation, provides improved performance with respect to the baseline. We experimentally demonstrate that our approach outperforms the baseline in terms of its ability to adapt to external perturbations, overall controllability provided to the subject, and feedback from participants on their perceived controllability. The amputee performed similarly to the able-bodied participants, indicating that the proposed framework may provide improved performance for the target population of transradial amputees.
Type of Work: | Thesis (Doctorates > Ph.D.) | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Award Type: | Doctorates > Ph.D. | ||||||||||||
Supervisor(s): |
|
||||||||||||
Licence: | Creative Commons: Attribution-Noncommercial-No Derivative Works 4.0 | ||||||||||||
College/Faculty: | Colleges (2008 onwards) > College of Engineering & Physical Sciences | ||||||||||||
School or Department: | School of Computer Science | ||||||||||||
Funders: | Engineering and Physical Sciences Research Council | ||||||||||||
Subjects: | Q Science > Q Science (General) | ||||||||||||
URI: | http://etheses.bham.ac.uk/id/eprint/14136 |
Actions
Request a Correction | |
View Item |
Downloads
Downloads per month over past year