Evolving Character Controllers for Collision Preparation

m.s. thesis, 9/2006, oregon state university
robert rose, under the guidance of dr. ron metoyer

abstract

    Full dynamic control of physically simulated characters is the holy grail of video games and other interactive applications. Recent advances in the field have presented controllers that can handle the balance and fall response after a collision. This thesis addresses the development of a controller that handles the response before a collision--reflexive posturing that protects the character from impact. To solve this problem we present a technique that uses genetic algorithms to optimize a pose controller based on analysis from a pain measurement system. Applying our technique we develop collision preparatory controllers that defend a virtual character in several different styles from a variety of threats.
thesis related paper
    (To appear) Zordan, Macchietto, Medina, Soriano, Wu, Metoyer, Rose. Anticipation from Example. In Proceedings of the ACM Symposium on Virtual Reality Software and Technology, November 2007. [PDF]