Neural Driven Gait Synthesis with Temporal Convolutional Networks


Balint Hodossy


Imperial College London, CDT in Prosthetics and Orthotics

Document Type


Developing devices that support and restore mobility is an increasingly important objective,
due to the projected demographic shift towards an aging population. Actuated devices, like
exoskeletons, aim to facilitate activities of daily living such as climbing stairs or walking
independently in cases where energetically passive devices are not sufficient. However,
unintuitive interfacing and control strategies reduce the number of cases where these devices
could be a worthwhile intervention.
This study investigates motion synthesis models that predict body segment kinematics from
trunk muscle surface electromyography (sEMG) processed with temporal convolutional
networks (TCN) to establish a voluntary, hands-free, and responsive control interface. Both a
direct regression and an abstracted, hierarchical control approach are explored.


Leave a comment:
Rate UsPoorFairGoodVery goodExcellentRate Us

Thanks for submitting!