Soft Robotics: Cerebellar Inspired Control of Artificial Muscles

Soft robots have the potential to greatly improve human-robot interaction via intrinsically safe, compliant designs. However, new compliant materials used in soft robotics – artificial muscles – are fabricated with poor tolerances and have time-varying dynamics. Therefore, a key technical challenge is to develop adaptive control algorithms for these materials. Here, we take a novel bioinspired approach to artificial muscle control using the adaptive filter model of the cerebellum. The cerebellum is a brain structure essential for fine-tuning human performance in a diverse range of sensory and motor tasks. Its ability to automatically calibrate and adapt to changes in a wide variety of systems using a homogenous, repeating structure suggests that cerebellar-inspired models are highly suited to controlling artificial muscles in a range of tasks. We investigate the performance of the cerebellar adaptive filter algorithm in the displacement control of a soft actuator. Experimental results demonstrate that the cerebellar algorithm is successful and learns to accurately control the time-varying dynamics of the soft actuator in real-time.