Abstract:
The primary purpose of a computer animation system is to provide assistance to the human animator in synthesizing the movement of the animated character such that the resulting motion appears physically correct and natural. If the character is modelled as an articulated figure consisting of an appropriate number of links and joints, then it's movement could be specified by a trajectory - time dependent values for the position of the figure. Hence all possible trajectories for a given set of goals to the character can be considered as forming a space of trajectories and the problem of synthesizing movement can be viewed as that of searching in this space for a trajectory satisfying all the physical requirements and the animator's goals. Unfortunately, this space is normally of very high dimension, and even for relatively simple characters and goals, the solution space is vast, multimodal and discontinuous, not lending itself very well to the use of conventional gradient based search techniques. This paper presents an evolutionary programming technique that uses motion features to direct the search. While genetic algorithms for motion synthesis have been proposed by other researchers, the computation of motion features and their use in the definition of a fitness function for determining subsequent generations is novel. The method has been implemented and simulation experiments with characters of reasonable complexity have yielded very good results. In this paper, we describe in detail the evolutionary programming technique that we have devised, computation of motion features and fitness function formulation. We also present results from some of the experiments conducted using our implementation.
