Composable Controllers for a Virtual Stuntman

The implementation of a multi-talented, dynamic virtual actor is a significant challenge given the difficulty of designing controllers even for very specific motions. Divide-and-conquer methods are a logical approach to this problem, but this requires a community of control practitioners as well as a means of integrating the fruits of their labor. We propose a framework for composable controllers to address the latter requirement, which in turn is also an important prerequisite for the former. The composition framework is tested on two novel families of controllers. The first is aimed at the basic motions which need to be at the core of any skill repertoire, such as balance and fall recovery. The second explores a practical application of a synthetic actor, namely the execution of dangerous stunts by a dynamic digital stuntman.