Animating Human Finger Movements: A Feedforward Neural Network Applied to Artificial Tendons

O.A. Kuchar and J.N. Scrimger

Faculty of Computer Science, DalTech - Dalhousie University
P.O. Box 1000, Halifax, N.S., Canada, B3J 2X4
{okuchar | scrimger}


In our research, we are focusing on the preliminary developments of creating an artificially intelligent virtual human (AIVH) who is capable of moving and interacting in its environment with little intervention from the animator. Before an AIVH can move completely, an AIVH needs to learn how to move its body. Thus, the body needs to be divided into separate areas and the AIVH needs to learn about movements in these areas. In this paper, we report the use of a feedforward artificial neural network (ANN) to animate a computerized human finger. The underlying model of the computerized human finger is based on a biological model involving tendons. The ANN triggers any tendons that need to be manipulated to achieve an animated goal. The prescribed animated goals involve different flexion and extension movements. There are several drawbacks associated with using a multilayer feedforward network. Firstly, since the architecture is designed by trial-and-error, it is very difficult to create an effective a priori architecture. Secondly, ANNs are not trained on an entire input space, but there is a supposition that the ANN will work correctly when presented with any possible input within this problem space. In this ongoing project, the ability for an AIVH to move its finger using an ANN has been developed and tested, producing positive results.

Keywords: biological-based animation, artificial neural networks, backpropagation training algorithm, motion control, human figure movements