ABSTRACT
The need for a well-informed steering solution and total freedom for each team to implement animation state graphs that suited their needs called for an unconventional solution to the problem of accurate steering and path following. Our solution treats the game as a black-box system, learning the character's movement parameters to later build an independent motion model that informs steering. This chapter presents a way of significantly decoupling dependencies between AI and animation. This approach treats animation as the dominant party in this relationship of motion and shows a way of gaining benefit from such an approach. The usage of simulations to improve steering control of an animation-driven agent has proved to be a success. Our desire is to be able to construct an arbitrarily complex motion model that steering can use to accurately control actors at any given desired velocity or angle, moving an actor to the goal without ever accidentally leaving the navigation mesh.
