Bioinspired Adaptive Control for Artificial Muscles

The new field of soft robotics offers the prospect of replacing existing hard actuator technologies by artificial muscles more suited to human-centred robotics. It is natural to apply biomimetic control strategies to the control of these actuators. In this paper a cerebellar-inspired controller is successfully applied to the real-time control of a dielectric electroactive actuator. To analyse the performance of the algorithm in detail we identified a time-varying plant model which accurately described actuator properties over the length of the experiment. Using synthetic data generated by this model we compared the performance of the cerebellar-inspired controller with that of a conventional adaptive control scheme (filtered-x LMS). Both the cerebellar and conventional algorithms were able to control displacement for short periods, however the cerebellar-inspired algorithm significantly outperformed the conventional algorithm over longer duration runs where actuator characteristics changed significantly. This work confirms the promise of biomimetic control strategies for soft-robotics applications.