A novel, neuroscience-based control paradigm for wearable assistive devices : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Engineering at Massey University, Albany, New Zealand
The biological domain has evolved in such a way that it effciently overcomes many
problems we struggle to solve in the engineering domain; for example, bipedal locomotion,
which requires a number of desirable attributes, e.g. compliance and
adaptability. As such, the aim of this research has been to provide a bridge between
the biological and engineering domains, capturing these attributes, and developing
an enabling control technology. The application of this research has been around
wearable assistive devices: devices that assist rehabilitation and recuperation of lost
or impaired functions or enable an end user to perform difficult to complete tasks.
As such, this thesis presents a novel, neuroscience-based control technology for wearable
assistive devices. Major contributions of this work include reproducing both
biological movement's compliant and adaptive properties in the engineering domain.
The presented approach consists of using an assistive device, whose joints are
antagonistically actuated using compliant pneumatic muscles, and central pattern
generators. The assistive device's actuators make the arm robust to collision and
give it smooth, compliant motion. The pattern generators produce the rhythmic
commands of the joints of the assistive device, and the feedback of the joints' motion
is used to modify each pattern generator's behaviour. The pattern generator enables
the resonant properties of the assistive device to be exploited to perform a number
of simulated rhythmic tasks.
As well as providing a wealth of simulated and real data to support this approach,
this thesis implements integrate-and-fire, Izhikevich, and Hodgkin-Huxley neuron
models, comparing their output based on ring patterns observed in neurons of
the nervous system. These observations can be used as a mechanism for deciding
the "realism" needed to represent a neural system's characteristics. In addition,
Hill's muscle model has been presented, and simulation of an implemented soleus
muscle carried out. Parametric variation provides quantitative insight into passive
and active series and parallel elements' roles in generating tension and tension's timeresponse
characteristics. Furthermore, an antagonistically coupled pair of extensor
flexor muscles have been presented and shown to effect compliant joint actuation
of a modelled limb under differential activation. Co-activation of the extensor and
flexor has been shown to increase a joint's stiffness, leading to increased stability and
rejection of limb perturbation.