Page 319 - Handbook of Biomechatronics

Upper and Lower Extremity Exoskeletons 311

4.4 Control for Exoskeleton
Despite decades of research relating to multifunctional myoelectric control,
there is much to be made before myoelectric control can effectively be inte-
grated into daily life commercial applications. From the ability to extract
proper muscle activity information for most potential users, the future of
simultaneous multifunctional control applications relies on producing reli-
able control schemes utilizing robust representations of muscle synergies
(Ison and Artemiadis, 2014). The future directions should lie in three main
areas: the development of real-time control applications and standardized
metrics to compare performance across differing techniques and improve
the surface electromyography recordings through high-density surface
EMG (HDsEMG) and the development of hybrid prediction and learning
schemes for user-friendly control (Ison and Artemiadis, 2014).
The motor learning-based control schemes train a motor system to
develop and refine synergies associated with system dynamics of a specific
mapping function relating sEMG inputs with control outputs. The user
learns the system dynamics via feedback while interacting with the control
interface. This scheme consistently reports significant learning while achiev-
ing good performance metrics (Ison and Artemiadis, 2014). These metrics
are generally specific to the given task and are difficult to compare to other
control methods implemented in real time. For real-time implementation
and testing of control schemes, it is necessary to standardize metrics that
can compare performance and efficiencies across different schemes, includ-
ing comparisons between pattern recognition and motor learning (Ison and
Artemiadis, 2014).
Advancements in recording technology have made HDsEMG electrodes
a viable option for myoelectric controllers. The high-density electrodes pro-
vide a more complete set of information to allow for richer processing and
more robust control schemes (Ison and Artemiadis, 2014). From a macro-
scale view, HDsEMG provides opportunities to describe two-dimensional
distributions of muscle activity as well as intensity, compensating for elec-
trode shift and cross talk. In addition, provides redundancy in signals such
that they can be subsets that allow for more efficient estimation without
losing control performance (Ison and Artemiadis, 2014).
Attempts to reduce the training phases have been made in classification
schemes using adaptive learning and pretrained models. The hybrid
approach, which aims to use natural population-wide approaches in order
to develop new forms of synergies, may be the key to efficient, user-friendly