Analytical and Deep Neural Network Based Hybrid Modelling of Single-Segment Continuum Robot

Abstract

Hyper-redundancy and ability to propagate through complex curvilinear pathways have made the continuum robots immensely popular in different fields of application. Precise control of continuum robot position relies on the accurate knowledge of the manipulator kinematics. However, the inverse kinematic model of continuum robots based on analytical modelling fails to include the unmodeled dynamics and effect of disturbances due to gravity and friction resulting in low accuracy in modelling and control. On the other hand, purely data driven continuum models require huge amount of data and time for training and suffer from high inaccuracy in modelling as well. Hybrid modelling in this scenario has been perceived to achieve better accuracy by combining the advantages of these two modelling approaches. Therefore, in the present work, we focus on the inverse kinematics of the single-segment continuum robot by using the analytical and deep neural network based hybrid modelling approach. Experimental results indicate that the proposed method increases the modelling accuracy and also at the same time it reduces the network training time significantly.