Pattern recognition-based real-time myoelectric control for anthropomorphic robotic systems : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Mechatronics at Massey University, Manawatū, New Zealand
Advanced human-computer interaction (HCI) or human-machine interaction (HMI) aims to help
humans interact with computers smartly. Biosignal-based technology is one of the most promising
approaches in developing intelligent HCI systems. As a means of convenient and non-invasive
biosignal-based intelligent control, myoelectric control identifies human movement intentions from
electromyogram (EMG) signals recorded on muscles to realise intelligent control of robotic systems.
Although the history of myoelectric control research has been more than half a century, commercial
myoelectric-controlled devices are still mostly based on those early threshold-based methods. The
emerging pattern recognition-based myoelectric control has remained an active research topic in
laboratories because of insufficient reliability and robustness. This research focuses on pattern
recognition-based myoelectric control. Up to now, most of effort in pattern recognition-based
myoelectric control research has been invested in improving EMG pattern classification accuracy.
However, high classification accuracy cannot directly lead to high controllability and usability for
EMG-driven systems. This suggests that a complete system that is composed of relevant modules,
including EMG acquisition, pattern recognition-based gesture discrimination, output equipment and its
controller, is desirable and helpful as a developing and validating platform that is able to closely emulate
real-world situations to promote research in myoelectric control.
This research aims at investigating feasible and effective EMG signal processing and pattern
recognition methods to extract useful information contained in EMG signals to establish an intelligent,
compact and economical biosignal-based robotic control system. The research work includes in-depth
study on existing pattern recognition-based methodologies, investigation on effective EMG signal
capturing and data processing, EMG-based control system development, and anthropomorphic robotic
hand design. The contributions of this research are mainly in following three aspects:
Developed precision electronic surface EMG (sEMG) acquisition methods that are able to
collect high quality sEMG signals. The first method was designed in a single-ended signalling
manner by using monolithic instrumentation amplifiers to determine and evaluate the analog
sEMG signal processing chain architecture and circuit parameters. This method was then
evolved into a fully differential analog sEMG detection and collection method that uses
common commercial electronic components to implement all analog sEMG amplification and
filtering stages in a fully differential way. The proposed fully differential sEMG detection and collection method is capable of offering a higher signal-to-noise ratio in noisy environments
than the single-ended method by making full use of inherent common-mode noise rejection
capability of balanced signalling. To the best of my knowledge, the literature study has not
found similar methods that implement the entire analog sEMG amplification and filtering chain
in a fully differential way by using common commercial electronic components.
Investigated and developed a reliable EMG pattern recognition-based real-time gesture
discrimination approach. Necessary functional modules for real-time gesture discrimination
were identified and implemented using appropriate algorithms. Special attention was paid to
the investigation and comparison of representative features and classifiers for improving
accuracy and robustness. A novel EMG feature set was proposed to improve the performance
of EMG pattern recognition.
Designed an anthropomorphic robotic hand construction methodology for myoelectric control
validation on a physical platform similar to in real-world situations. The natural anatomical
structure of the human hand was imitated to kinematically model the robotic hand. The
proposed robotic hand is a highly underactuated mechanism, featuring 14 degrees of freedom
and three degrees of actuation.
This research carried out an in-depth investigation into EMG data acquisition and EMG signal pattern
recognition. A series of experiments were conducted in EMG signal processing and system
development. The final myoelectric-controlled robotic hand system and the system testing confirmed
the effectiveness of the proposed methods for surface EMG acquisition and human hand gesture
discrimination. To verify and demonstrate the proposed myoelectric control system, real-time tests were
conducted onto the anthropomorphic prototype robotic hand. Currently, the system is able to identify
five patterns in real time, including hand open, hand close, wrist flexion, wrist extension and the rest
state. With more motion patterns added in, this system has the potential to identify more hand
movements. The research has generated a few journal and international conference publications.