Motion blending is commonly thought of as creating the transition of an animated figure to the first frame of a piece of motion. We describe a new way of thinking about motion blending by removing the assumption that each piece of motion must start at the first frame. We describe a classifier system that finds the nearest match in a set of animation frames to a given state of an animated figure and show how this classifier can be used to create better motion blends. We also describe how the parameters for this system were optimised using a genetic algorithm.

The classifier given was found to be efficient enough to work in real time with many articulated figures.