Bio-inspired robotic structure composed of soft actuation units has attracted increasing research interests in its potential and capacity of complying with unstructured and dynamic environment, as well as providing safe interaction with human; however, this inevitably poses technical challenging to achieve steady, reliable control due to the remarkable non-linearity of its kinematics and dynamics. To resolve this challenge, we propose a novel control framework that can characterize the kinematics of a soft continuum robot through the hyper-elastic Finite-element modeling (FEM). This enables frequent updates of the Jacobian mapping from the user motion input to the end-effector's point of view. Experimental validation has been conducted to show the feasibility of controlling the soft robot for intracavitary path following. This could be the first success to demonstrate the perspectives of achieving stable, accurate and effective manipulation under large change of robot morphology without having to deduce its analytical model. It is anticipated to draw further extensive attention on resolving the bottleneck against the application of FEM, namely its intensive computation.