An Artificial Neural Network Based Haptic Rendering Of Contact With Deformable Bodies

This paper presents an artificial neural network based 3-DOF haptic rendering scheme to render the contact force between a rigid object and a deformable body in a virtual environment. The finite-element method (FEM) technique is widely used for solving the deformation problem. However, this method has a heavy computational load to get accurate result, so it is difficult to apply this method to haptic simulating. To solve the challenging problem, in this paper, we presented a new motion control scheme to divide the motion of a rigid virtual object into three sub-movements along the three axes of a Cartesian-coordinate, based on which three single-input and single-output neural networks can be separately used to compute the three feedback force components along all the coordinate axes. The vector composition of the three force components is the feedback force exerted to a user through a haptic device. The proposed method can ensure the high accuracy and the high update rate of 3-DOF haptic rendering of deformable bodies. To testify the accuracy of the artificial neural network for haptic rendering, a medical robot is used to measure the data of the neural network training in the physical world, and a haptic device based experiment with a virtual environment validates the proposed algorithm for 3-DOF haptic rendering.