Modeling soft robotic actuators using data-driven model reduction

Abstract

Soft robotics is a emerging research area in the field of robotics with promising application areas where traditional robots fails to perform efficiently. However obtaining a nearly accurate mathematical model of the soft robot is challenging due to its high flexibility and non-linear material properties. The limited information of the system in terms of its parameters have led to alternative modeling approaches. With large scale data collection in terms of evolution of states and system identification, data-driven techniques are favorable alternative to obtain nearly accurate models. Considering the large scale dimensional problem of the resulting model, model reduction technique have been applied to obtain optimal reduced set of governing equations. In this work, the dynamic mode decomposition based system identification technique have been implemented in the soft robot model. This technique does not need any information regarding the parameters of the system and is solely dependent on the measurement of the system states. The obtained dynamic model is a linear approximation of the full-order system. The paper describes the model reduction technique aligned with the soft robotic actuator application.