Data-driven bending angle prediction of soft pneumatic actuators with embedded flex sensors

In this paper, resistive flex sensors have been embedded at the strain limiting layer of soft pneumatic actuators, in order to provide sensory feedback that can be utilised in predicting their bending angle during actuation. An experimental setup was prepared to test the soft actuators under controllable operating conditions, record the resulting sensory feedback, and synchronise this with the actual bending angles measured using a developed image processing program. Regression analysis and neural networks are two data-driven modelling techniques that were implemented and compared in this study, to evaluate their ability in predicting the bending angle response of the tested soft actuators at different input pressures and testing orientations. This serves as a step towards controlling this class of soft bending actuators, using data-driven empirical models that lifts the need for complex analytical modelling and material characterisation. The aim is to ultimately create a more controllable version of this class of soft pneumatic actuators with embedded sensing capabilities, to act as compliant soft gripper fingers that can be used in applications requiring both a ‘soft touch’ as well as more controllable object manipulation