Intent-Driven Gait Synthesis for Simulated Assistive Devices

The rapid advancement of BMI technology has reached the point of clinical studies; however,
further development is hindered due to a need to rely on invasive implants for high quality,
information carrying signals. In order to increase the number of cases where BMIs can provide
worthwhile treatment for a user, less risky, non-invasive methods need to be investigated. Neural
networks have been shown to be effective at processing neural signals, however, many complex
architectures using multi-modal inputs are yet to be investigated. My proposed project aims to
address these needs in three parts. First, by applying a neural network to decode the underlying motor neuron activity from
surface signals.
Second, to investigate different machine learning approaches which can extract the intent
from these signals, and reliably control a high degree of freedom prosthesis or orthosis.
Lastly, to integrate these processes in a system which learns together with the user; giving
feedback on the signals they provide and adapting to their unique needs. Meanwhile the user
learns how to efficiently live with their device.
If this system is successfully implemented, it could eliminate the need for professional supervision of
the calibration of BMIs and increase the ability and confidence of users to act independently.