Optimal Motion Control using a Wavelet Network as a Tunable Deformation Controller

Several motion control techniques rely on shape control rather than actuator command. A shape controller speciies the desired shape of the object, but not its position and orientation. The actual position and orientation result from interaction with the surrounding world. We propose to use a bank of wavelet networks, together with local optimization algorithms, as a tunable deformation controller. This enables to represent locomotion gait by a set of a few signiicant parameters, which are the wavelet node weights, dilatation and translation parameters. These parameters can be extracted from motion analysis. They can in turn be applied to control a synthetic creature. Local optimization of the wavelet network parameters make it feasible to fullll user-speciied goals, while preserving the general features of the creature locomotion gait. This approach has been sucessfully applied to the motion control of a synthetic whale in open-loop.