Learning a Curvature Dynamic Model of an Octopus-inspired Soft Robot Arm Using Flexure Sensors

Abstract

This work presents a novel technique for sensing body dynamics for a soft-robot arm inspired by the octopus, using flexure sensors. The aim is to develop a sensing technique which can also simultaneously enable learning of the body dynamics that can then be used for control. Flexure sensing is advantageous for a soft bodied robot since it is a direct measure of local behaviour along the arm, and is closely connected with the piecewise constant curvature assumption employed for such robots. Initial results on simulated sensor measurements and dynamics learning are presented and ongoing work and applications are discussed.